My brother is a tradesperson. If I filled my garage with the same DeWalt drills, saws, laser levels and every other tool he owns, I’d have a very impressive workshop. I still wouldn’t know what he knows.

He’ll look at a frame and immediately see what’s wrong. He’ll stop halfway through a job because something doesn’t feel right. He’ll spend an extra hour fixing something no client will ever notice because he knows they’ll notice in five years.

None of that lives in the toolbox.

Subscribe now

I’ve been thinking about that while watching the race around AI coding assistants. Every week there’s another model, another benchmark and another video showing software appearing almost magically on the screen.

It’s impressive.

It’s also easy to mistake the tool for the capability.

We’ve done this before. Spreadsheets didn’t produce better financial decisions. CAD didn’t produce better buildings. PowerPoint didn’t produce better strategy. Those tools made capable people more effective, but they never replaced experience, judgment or accountability.

AI feels like the next version of the same story.

Writing code is becoming easier. That’s remarkable. Deciding what should be built, why it matters and what happens next is still the hard part. In fact, as execution gets cheaper, those decisions probably matter even more.

That’s the part I think we sometimes miss.

When everyone has access to extraordinary tools, the advantage shifts somewhere else. It shifts to judgment. To taste. To knowing when not to build something. To recognising the difference between something that works and something that lasts.

My brother has never talked much about his tools. He talks about the job, the materials, the client and the finish. The tools are just there, quietly doing what they’re supposed to do.

Maybe that’s how we’ll think about AI one day.

Not as the thing that made great builders.

Just another tool that helped them build.

What do you think?

Leave a comment

She works with B2B companies on growth, and those companies spend thousands each year on CRM licences, data-enrichment tools and people who research and verify contact information. The database is not a mailing list bought on the cheap. It is part of how they compete.

Then HubSpot sent an email to administrators about changes to its enrichment products.

The email said that from 4 August 2026, enrichment data such as business contact details, employer information and email-deliverability signals “may be shared with other customers”. Customers could stop this by changing their settings before the deadline.

Chotai’s reaction was blunt:

HubSpot co-founder and CTO Dharmesh Shah replied in the comments:

It is hard to improve on that response. No throat-clearing. No claim that customers had misunderstood the technology. No paragraph drafted by a committee explaining that the company remained deeply committed to trust.

“Sorry. You are right.”

HubSpot reversed the decision within days. In its fuller response, the company said it would not proceed with the terms changes and that any future version of the enrichment scheme would be presented on an opt-in basis.

Subscribe now

The apology was good. The decision that prompted it is more interesting.

What did HubSpot think it was buying?

HubSpot’s logic is easy enough to reconstruct.

A large pool of business-contact data can make an enrichment product more accurate. One customer corrects a job title or validates an email address; another customer gets a better record. The data becomes more useful as more companies participate.

That sounds sensible when described as a product feature.

It sounds rather different when described from the customer’s side.

A company pays for HubSpot. It pays for data services. It pays employees to find the right people, verify their details and record what they learn. After years of work, the CRM contains a map of the company’s market: who it knows, who it is pursuing, which accounts matter and where opportunities are moving.

HubSpot’s email referred to business contact details, employer information and deliverability signals. Those fields may look ordinary in isolation. Their value often comes from the work around them.

A person’s work email might be public. The fact that your company identified that person, connected them to a target account, verified the address and decided they were relevant to a buying process is not quite the same thing.

That is why “your data belongs to you” does not settle the matter.

A vendor can leave ownership of the original data untouched while gaining the right to extract signals from it, combine those signals with other customers’ records and use the result to build a better commercial product.

The customer still owns the rows in the database. The vendor owns the thing made from them.

Share

A CRM is not a social network

We have spent two decades living with the social-media bargain.

The service costs nothing in cash, so the platform makes money from attention and data. Most users understand the broad outline, even if few read the terms or grasp the extent of the tracking.

That bargain does not transfer neatly to enterprise software.

A business using HubSpot is already paying. It has probably also paid for implementation, integrations, consultants and staff training. Once the platform sits at the centre of sales and marketing, replacing it becomes expensive and disruptive.

The company did not sign up thinking, “We get the software, and in return the vendor gets to learn from our commercial relationships.”

That is the key difference.

A social-media user may suspect that their activity is part of the business model. A SaaS customer expects the subscription fee to be the business model.

When a vendor later finds a second source of value in the customer’s data, it is changing the economics of the relationship. Doing that after the data has been loaded and the workflows have been built makes the change harder to refuse.

An opt-out setting does not remove that problem. It puts the burden on the customer to notice the change, understand it, find the correct control and act before a deadline.

In a large company, the email may go to a system administrator rather than the legal or procurement team. The administrator may have the technical permission to change the setting without having the authority to license the company’s data for a new purpose.

Slack had its own version

Slack ran into a similar argument in 2024.

Customers discovered that Slack could use workspace data to improve what it called global machine-learning models. Organisations that did not want to participate had to opt out by contacting Slack.

Slack said these models supported features such as search and recommendations. It drew a distinction between conventional machine learning and generative AI, and its privacy principles state that it will not use customer data to train generative AI models without affirmative opt-in consent.

That matters to Slack’s engineers and lawyers. It may matter less to the customer.

An internal Slack workspace contains product discussions, commercial plans, staffing decisions, customer issues and the informal exchange through which much of a company’s work gets done.

The obvious question was not whether Slack was training a large language model.

It was why the default allowed private workplace activity to help improve models used across Slack’s service.

Slack had access to the conversations because it had been hired to carry them. Customers did not automatically see that as permission to learn from them for a wider purpose.

The opt-out process made the problem worse. A company had to know the practice existed before it could object to it.

Share For Every Scale

Adobe found the same boundary in creative work

Adobe’s row arrived through an update to its terms of use.

The language gave Adobe broad rights to access and analyse customer content. Designers, photographers and other creative professionals quickly asked whether work stored or processed through Adobe products could be used to train Firefly, Adobe’s generative AI system.

Adobe said it did not train generative AI on customer content and rewrote its terms to make that commitment clearer. Its response included a simple assurance: “You own your content.”

Again, ownership was only part of the concern.

Creative professionals had paid Adobe for tools. They had not knowingly offered their finished work, drafts or client material as training material for the next generation of those tools.

Adobe’s problem came from language broad enough to cover more than customers expected. Once generative AI entered the picture, old phrases about accessing content and improving services acquired a much larger meaning.

Ten years ago, “improving the service” might have suggested bug fixes, performance monitoring and usage statistics.

Today it may include training models, extracting patterns and creating assets that grow more valuable with every customer contribution.

That change has made a lot of standard SaaS language look badly out of date.

The mistake these companies keep making

HubSpot, Slack and Adobe sell different products and handle different kinds of information.

The common mistake is assuming that access granted to deliver the service also creates permission to pursue a related business opportunity.

The contract may allow it. The privacy policy may mention product improvement. A setting may give the customer some control.

None of those things guarantees that customers will consider the use fair.

Customers tend to draw a practical line:

Use the data to provide the product we bought. Secure it. Back it up. Fix errors. Make the service work better for us.

Ask before using it to build something for everyone else.

Vendors often see no clear technical break between those activities. Data flows through the same systems. Models can improve search, recommendations, enrichment and automation. Every improvement can be described as helping customers.

The commercial break is much clearer.

When one customer’s data improves a product sold to another customer, the first customer has contributed an asset. The contribution may be small. Across thousands of customers, it can be enormously valuable.

That is why this is a procurement issue as much as a privacy issue.

What buyers should ask before the data goes in

General counsels will usually find reassuring language saying the customer retains ownership of its data.

They should keep reading.

The useful questions concern what the supplier may do with the data, what it may derive from it and whether those derived assets survive after the customer leaves.

Can the supplier use customer data to improve a model used by other organisations?

Does that include only raw data, or also metadata, prompts, outputs, embeddings, labels and usage patterns?

Can the supplier retain what it has learned after the source data is deleted?

Can the rules change through an online policy update?

Who inside the customer’s organisation can switch these uses on?

Chief procurement officers should ask the same questions before negotiating the price.

A supplier asking for broad learning rights is asking for something of value. That should not disappear into boilerplate beneath the subscription fee.

There is nothing inherently wrong with a shared-data product. Customers may choose to participate because they receive better information in return. Some may see genuine value in a network that improves as its members contribute.

The choice needs to be clear, informed and made before participation begins.

“Switch this off by next month” is a poor substitute.

Leave a comment

HubSpot’s apology deserves credit

Technology companies are rarely short of language when they need to avoid saying they were wrong.

Dharmesh Shah used thirteen words:

That response worked because it dealt with the complaint the customer had actually made.

Chotai was not asking for a technical explanation of enrichment. She was saying that HubSpot had crossed a line. Shah agreed and moved the line back.

Other SaaS leaders should study the response. They should also ask why a customer had to raise the alarm in the first place.

The lesson is not that companies should stop using data to improve software. Modern products cannot operate that way.

The lesson is that “improve our products and services” no longer provides enough information. Buyers need to know whose product is being improved, whose data pays for the improvement and who owns the result.

HubSpot customers thought they were building their CRM.

They did not expect to be building HubSpot’s next data product at the same time.

That was the deal customers rejected.

That moment has arrived.

Subscribe now

The market has shifted noticeably over the past six months. Microsoft has introduced usage-based billing for advanced Copilot capabilities such as Copilot Cowork alongside administrative budgets and spending controls. GitHub Copilot has moved to an AI credit model for premium usage. OpenAI and Anthropic continue to expand premium reasoning models that are priced according to consumption rather than simple subscription tiers.

This isn’t simply a pricing story. It’s a change in the economics of enterprise AI.

One of the better descriptions of what’s happening comes from Artefact, which describes the phenomenon as the “token cost illusion.” Individual token prices continue to fall, yet enterprise AI bills continue to grow because organisations are consuming far more compute than they were a year ago. Larger context windows, agentic workflows, multiple model calls and tool orchestration all increase consumption, often by more than the underlying price reductions offset.

That combination changes executive behaviour.

For the past two years success was measured by adoption. Organisations wanted employees experimenting with AI, learning new tools and incorporating them into everyday work. Today a different set of questions is appearing in executive meetings. Which teams genuinely need frontier models? Which work justifies autonomous agents? Which activities can be handled perfectly well by smaller, cheaper models?

Those questions aren’t about software procurement. They’re about investment.

From Universal Access to Deliberate Allocation

Enterprise technology eventually became standard equipment. Laptops, smartphones, collaboration platforms and productivity software all followed a similar pattern. Costs declined, deployment expanded and broad access became the obvious choice.

AI is developing differently because capability and cost remain tightly connected. A lightweight assistant, a frontier reasoning model and an autonomous agent deliver different levels of performance and consume very different amounts of compute. The assumption that everyone should receive the same capability is becoming much harder to defend.

I’ve started seeing this play out in conversations with industry peers. Premium AI licences are being prioritised rather than distributed automatically. Business units are being asked to explain the value they expect to create before receiving access to more capable models. Engineering, finance, legal, marketing and customer operations can all present persuasive cases. Few organisations have budgets that allow them to approve every request.

This isn’t evidence that organisations are losing confidence in AI. If anything, it demonstrates the opposite. Businesses apply governance to the capabilities they believe will matter most.

Share

A Different Kind of Capital Allocation

Recent research from KPMG helps explain why these discussions are becoming more common. Only 35 per cent of organisations report having full visibility into their AI operating costs, yet organisations with strong cost visibility are five times more likely to report established returns from their AI investments.

Those numbers are revealing because they shift attention away from technology and towards management. Executives cannot make sensible investment decisions without understanding the economics of the resource they are allocating.

Grant Gross captured the same transition in a recent CIO article, writing that “the AI adoption spending spree is over. Time to focus on value.” The article also describes organisations introducing controls after AI budgets were consumed much faster than expected, including Uber placing limits on AI coding tools after annual budgets were exhausted in only a few months.

These examples point to the same conclusion. Enterprise AI is moving away from broad deployment and towards deliberate allocation.

AI Budgets Are Becoming Strategy Documents

Budgets have always reflected organisational priorities. AI budgets reveal something slightly different. They indicate where leadership believes better judgement, faster analysis and higher-quality decisions are likely to create the greatest commercial return.

Organisations can invest similar amounts in AI and produce very different outcomes because they strengthen different parts of the business. One may concentrate on software engineering. Another may focus on customer operations. A third may invest most heavily in finance, legal or research. The technology is broadly similar, but the pattern of investment reveals a fundamentally different strategy.

That observation leads to a simple conclusion. The organisations that gain the greatest advantage from AI are unlikely to be those spending the most. They are more likely to be those making better decisions about where advanced AI capability creates disproportionate value.

Share For Every Scale

Executive Playbook: Allocating Intelligence

Recognising the shift is only the beginning. Executive teams also need a practical way to decide where additional AI capability belongs.

1. Start with business value

Begin with the outcomes that matter most to the organisation rather than the technology itself. Revenue growth, customer experience, operational efficiency, decision quality and risk reduction provide a much stronger starting point than discussions about models or licences. AI investment should support strategic priorities rather than create new ones.

2. Identify high-leverage work

Not every activity deserves the most capable AI available. The strongest business cases usually combine complex judgement, high frequency and meaningful commercial impact. Those activities generate returns that justify premium capability. Routine work often does not.

3. Match capability to the task

Different work requires different levels of intelligence. Many activities can be completed effectively using lightweight assistants or specialised models, while others benefit from frontier reasoning or autonomous agents. Organisations already apply this principle when assigning people with different experience and expertise to different kinds of work. AI should be managed in much the same way.

4. Measure outcomes, not activity

Usage statistics are easy to produce but rarely explain whether value has been created. Better measures include cycle time, customer outcomes, quality, financial performance and risk reduction. Those indicators provide a clearer basis for future investment decisions because they focus attention on business performance rather than technology adoption.

5. Review the allocation

Business priorities change. Technology changes. Markets change. AI investment should be reviewed with the same discipline applied to capital, talent and major programmes. Some teams will justify additional capability over time. Others may discover that less expensive tools produce comparable results.

Questions Worth Taking to the Executive Team

Every executive team should be able to answer five straightforward questions.

• Which decisions create the greatest value in our organisation?

• Where would better judgement or faster analysis produce the strongest commercial return?

• Which roles genuinely require frontier AI capability?

• How are we measuring business value rather than AI activity?

• If our AI budget doubled, or halved, how would our allocation change?

The discussion that follows those questions is likely to be more valuable than another debate about vendors, models or feature releases. Technology will continue to improve, prices will continue to move and new capabilities will continue to emerge. Those developments matter, but they are no longer the central management challenge.

Every generation of executives inherits a new strategic resource. Earlier generations learned how to allocate financial capital, digital technology, data and cloud infrastructure. This generation will learn how to allocate intelligence. The organisations that approach that task with the same discipline they apply to every other strategic investment are likely to discover that competitive advantage comes less from owning better AI than from putting it in the right places.

Leave a comment

References

Artefact. (2026). Is AI really getting cheaper? The token cost illusion. https://www.artefact.com/blog/is-ai-really-getting-cheaper-the-token-cost-illusion/

Gross, G. (2026). The AI adoption spending spree is over. Time to focus on value. CIO. https://www.cio.com/article/4183263/the-ai-adoption-spree-is-over-time-to-focus-on-value.html

KPMG. (2026). Growing adoption signals progress as cost visibility and accountability drive AI value. https://kpmg.com/xx/en/media/press-releases/2026/06/growing-adoption-signals-progress-as-cost-visibility-and-accountability-drive-ai-value.html

Rowe, J. (2025). 2026: The AI Invoice Arrives. For Every Scale. https://www.foreveryscale.com/p/2026-the-ai-invoice-arrives

I’ve realised that accountability is no longer the most important conversation CEOs need to have about AI. A bigger management challenge is emerging, and I suspect it will become one of the defining leadership issues of the next decade.

The conversation has changed

For the past two years, organisations have measured AI adoption by asking sensible questions. How many employees are using ChatGPT? How many Microsoft Copilot licences have we deployed? How many hours have we saved? How much productivity have we unlocked?

Those questions made perfect sense because AI was largely acting as another productivity tool. Employees used it to write documents, summarise meetings, analyse spreadsheets and brainstorm ideas. AI was helping people do their jobs better, but it wasn’t becoming part of the workforce itself.

That difference is beginning to disappear.

From tools to workers

I’ve noticed three separate conversations arriving at the same destination.

argues that AI is moving beyond chat interfaces towards agents that perform work. reaches a similar conclusion, describing agents as systems that pursue outcomes rather than simply responding to prompts. And, Mark Chatterton captured the management challenge in a single observation: “Day 1 was deployment. Day 2 is management.”

Taken together, they describe a much bigger shift than better software. They describe the beginning of a workforce that now includes autonomous digital workers operating alongside people.

That might sound like semantics, but I don’t think it is. Organisations know how to deploy software. They have far less experience managing thousands of autonomous systems making decisions, interacting with customers and completing work across multiple business processes.

Subscribe now

Management becomes the competitive advantage

Once AI starts performing work instead of simply assisting with work, the questions change completely.

Instead of asking whether employees are using AI, leaders need to understand where AI is operating, what authority it has been given and how its performance is being monitored. Those are no longer technology questions. They are management questions.

Think about how we manage people today. We recruit them, onboard them, grant access to systems, define their responsibilities, monitor performance, change their roles when required and eventually retire them from the organisation. Every one of those activities is supported by established processes, governance and management disciplines.

Now apply exactly the same thinking to AI agents.

Can you identify every agent operating across your organisation? Do you know what systems each one can access? Do you know who approved those permissions, how their performance is measured and who has the authority to stop them if something goes wrong? Those aren’t just technology questions. They’re questions about security, risk, governance and workforce management. If your CISO, Chief Risk Officer and CHRO aren’t already talking about AI agents together, they probably will be soon.

Share For Every Scale

The Bank of England is already thinking this way

The interesting thing is that regulators are beginning to have exactly this conversation.

Sarah Breeden, Deputy Governor of the Bank of England, argued that existing regulatory frameworks were not built for autonomous AI agents operating inside financial systems. Her concern was not chatbots producing inaccurate summaries or writing poor emails. It was autonomous systems making payments, executing trades and participating in financial markets at machine speed, where the traditional idea of a human reviewing every action starts to break down.

That moves the discussion beyond responsible AI principles and into the practical mechanics of management. How do you supervise autonomous systems? How do you intervene when they behave unexpectedly? How do you recover when something goes wrong? And who has the authority to stop them?

I think that’s an important signal for every executive, regardless of industry. One of the world’s leading central banks is no longer asking whether organisations should adopt AI. It is asking how organisations will manage, control and, when necessary, stop autonomous systems once they become part of everyday operations.

A new management discipline

Every major technology eventually creates its own management discipline.

The internet created digital marketing. Cloud computing created cloud operations. Cyber threats created cybersecurity as a board-level capability. Agentic AI will create something similar, although I don’t think we’ve settled on the name yet.

For now, let’s call it AI Workforce Management.

The label isn’t the important part. The important part is recognising that organisations are gradually building hybrid workforces made up of people and digital workers. The leaders who succeed won’t necessarily be those who deploy the most AI. They’ll be the ones who know exactly what their AI agents are doing, what authority they’ve been given, how they’re’re performing and when human intervention is required.

The next leadership question

When I wrote about accountability earlier this year, I argued that AI could never own responsibility for business outcomes. That remains my view, and I don’t think that principle changes simply because AI becomes more capable.

What has changed is the management challenge sitting above it. If the first question was who is accountable when AI decides?, the next question is who is managing the workforce that increasingly includes AI?

I suspect that will become one of the defining questions for CEOs over the next few years. The organisations that answer it well won’t just deploy AI more successfully. They’ll build an operating model that allows people and digital workers to contribute together, with the right governance, oversight and accountability sitting over both.

Share

Related reading

Time Under Tension newsletter

🤖 What is an agent?

One question we get asked more than any other right now: “What actually is an AI agent…

Read more

a month ago · 6 likes · Jason Ross

Day 1 was deployment. Day 2 is management by Mark Chatterton

One Useful Thing

The twilight of the chatbots

If you feel like things are accelerating in AI, you are probably right. Better AI models from the leading American AI labs have been releasing more quickly than ever (though government interventions stopped access temporarily to two of the most powerful models, Claude Fable and GPT-5.6…

Read more

25 days ago · 519 likes · 49 comments · Ethan Mollick

For Every Scale

Who Is Accountable When AI Decides?

AI is now performing parts of expert-level work in regulated industries…

Read more

5 months ago · 6 likes · Josh Rowe

The future state is always beautifully connected. Customer data flows into one platform. Journeys are mapped. Content is personalised. Dashboards show what worked. AI helps the team decide what to do next.

That is the dream.

The problem is that most organisations do not start there. They start with limited budgets, small teams, legacy systems, imperfect integrations, scattered data and more work than people.

That is why RMIT University is a useful case.

Not because every company is a university. It is useful because RMIT shows a more realistic pattern for AI adoption: make progress in chunks, learn fast, and stop waiting for the perfect system to arrive.

The perfect system is the trap

The phrase “right person, right message, right channel, right time” has been around long enough to lose most of its meaning.

It sounds sensible, but it hides the operating burden. To actually do it, a company needs data it can trust, journeys it can understand, channels that connect, content that can be produced, and measurement that shows whether anything changed.

That is a lot of machinery.

For leadership teams, the mistake is treating that machinery as something that must be solved all at once. The architecture diagram becomes the strategy. The integration plan becomes the excuse. The transformation program becomes so large that the business waits too long before anything useful changes.

Perfect integration sounds responsible. Often it is just a polite way to delay the first hard choice.

Constraints sharpen the question

RMIT’s context is relevant because it did not sound like a blank-cheque transformation story.

It had a big job to do and real constraints around money and people. That is the normal condition for most organisations. The work keeps growing, but the team and budget do not grow at the same speed.

That changes the adoption question.

A constrained team cannot afford to ask, “How do we build the perfect AI-enabled customer journey?” It has to ask, “Where can better data, better timing or better automation change an outcome now?”

That is a much more useful question.

It forces the team to look for leverage. Where is the journey too complex to manage manually? Where is the organisation guessing? Where would a more relevant intervention matter? Where can a small improvement be measured?

That is how AI adoption becomes practical.

RMIT gives the proof point

RMIT’s student journey is a useful example because it is long, messy and high-stakes.

Adobe’s case study says RMIT serves more than 90,000 students from over 190 countries, with journeys stretching from Year 10 through to final enrolment. It also points to disconnected touchpoints across open days, brochures and social channels.

That is not a simple funnel. It is a decision system.

The student is not always the only decision-maker. Parents matter. Geography matters. Study area matters. Timing matters. Confidence matters. Channel preference matters.

At Adobe Summit Sydney, RMIT talked about dozens of touchpoints and hundreds of segments. That is not just a marketing ambition. It is an execution burden.

The useful question is not whether AI can create more personalised content. The useful question is whether the organisation can understand the next few moments that matter, and act on them without overwhelming the team.

The first win was not magic

The official case study says RMIT implemented a data foundation in 15 weeks, improved website speed by 29%, and created a dashboard to track prospective student journeys across channels. It also says a personalised offer banner achieved a 23 per cent enrolment conversion rate.

Take the vendor framing with the usual caution. The sequence is still worth paying attention to.

The first useful work was not magic. It was foundation work.

Better first-party data. Better journey visibility. Faster web experience. A clearer view of where prospective students were engaging and where they were dropping off.

Then came a targeted intervention: a banner for students who had received an offer but had not yet enrolled.

That is a good example because the moment is specific. The audience is clear. The desired action is clear. The measurement is close enough to the behaviour to matter.

That is very different from vague personalisation at scale.

Early access can be a strategy

One underappreciated lesson from RMIT is that constrained organisations may need to be early, not late.

Being an early adopter is not just about chasing novelty. Used carefully, it can be a leverage strategy. It can give a smaller team access to capability, learning, vendor attention and influence it might not otherwise get.

There is a cost to that.

Early tools have rough edges. Integrations are imperfect. The team has to tolerate ambiguity while still running the business. Not every beta will be worth the time.

But waiting has a cost too.

By the time everything is mature, packaged and perfectly documented, better-resourced competitors may already have built the internal muscle. They may already know which workflows matter, which data is useful and which claims do not survive contact with the customer.

For constrained teams, the choice is not between perfect and imperfect. It is between learning now or waiting for certainty that may arrive too late.

Integration is never finished

The most honest part of the RMIT lesson is that martech integration is never really done.

That is not a criticism of one vendor. It is the nature of modern operating systems inside organisations. Every company has platforms, data sources, channels, approvals, legacy processes and teams that work around the official architecture.

AI does not arrive into a clean environment.

That means the practical question is not, “Is everything integrated?” It is, “Which chunk is valuable enough to improve next?”

This is where leadership discipline matters. Without discipline, the organisation either chases shiny tools or disappears into a multi-year integration program. Neither is ideal.

The better path is to pick a piece of work where the outcome matters, the data is usable, and the improvement can be measured. Then move to the next one.

The lesson for CMOs

RMIT’s story is not really about higher education.

It is about sequencing.

Most organisations will not build the perfect AI stack before they need to make progress. They will have to make progress while the stack is still messy, the integrations are still imperfect and the team is still learning what AI can actually change.

That is not a failure. It is the normal adoption path.

The CMO job is not to demand a perfect AI architecture before anything moves. It is to force a useful sequence.

Where does better data change a decision? Where does better timing change behaviour? Where does automation reduce manual work without reducing judgment? Where can the organisation prove value before scaling the pattern?

That is how constrained teams make AI useful.

Not by waiting for everything to connect.

By choosing the next useful chunk.

Disclosure: Adobe invited me to Adobe Summit Sydney and covered travel and accommodation. Adobe had no editorial review or approval. I’m interested in practical AI stories, not vague transformation language or paid placement. Pitch one here.

How do we show up in ChatGPT?

It is understandable. Customers are using AI tools such as ChatGPT, Perplexity, Gemini, Claude, and Copilot to search, compare, summarise and shortlist. The industry will argue over labels like AEO, GEO and AI search. I’m going to use AI search here because it is the clearest description of the business problem.

AI is becoming a new discovery layer.

Subscribe now

That means brands need to understand how they appear in AI-generated answers. But at Adobe Summit Sydney, Qantas made the more useful point.

The better question is not how to get mentioned by an AI tool. The better question is what the human behind the prompt is actually trying to decide.

That difference changes the whole conversation.

Visibility is not preference

The first wave of AI search advice will sound familiar. Track your mentions. Measure your citations. Optimise your content. Understand which prompts surface your brand. Make sure the model knows who you are.

Some of that will matter.

But showing up in an AI answer is not the same as being chosen. It is closer to making the consideration set. That is still useful, but it is not the whole game.

If a customer asks an AI assistant for a shortlist of airlines, insurers, universities, consultants, accounting platforms, CRM tools or project management software, a brand would rather be in the answer than outside it.

But the AI answer is not the customer.

The model may summarise the market. It may point the customer in a direction. It may collapse several searches into one answer. But the human still has to decide whether to trust what they have been given.

That is where the AI search story gets more interesting.

Qantas asked the better question

Gerrit Walters, Head of Design, Digital Experience & Performance for Qantas said the wrong question is to ask how to get mentioned more in AI Search. That is an inside-out question. It starts with the brand’s anxiety rather than the customer’s job.

The better move is to think about the person asking the question.

What are they trying to do? What would make their day easier? What evidence would help them decide? What would make them doubt the first answer? What else would they check before buying?

That is the part most AI search commentary is going to miss.

A customer does not suddenly stop being a customer because an AI tool helped them search. They still compare. They still check reviews. They still ask whether the brand is reliable, risky, premium, cheap, useful, credible or likely to disappoint them.

The AI answer may change the path.

It does not remove the human judgment at the end of it.

The brand moves into the context layer

Traditional digital marketing has spent years managing the content a brand publishes. Website pages. Landing pages. SEO copy. Product feeds. Campaigns. Social posts. Reviews, where the brand can influence them. Help content, where the brand remembers to maintain it.

AI search changes that because the answer is not just a link to your content. It is a synthesis of the context around your brand.

That context may include your website, reviews, forums, product documentation, press coverage, comparison pages, social posts, customer complaints, third-party rankings, creator commentary and whatever else the model can retrieve or infer.

That means the marketing job shifts.

It is no longer just about managing the message. It is about managing the evidence environment around the brand.

That is a much harder task. It is also a more honest one.

Share

AI search needs a measurement layer

There will be a rush to understand how brands appear in AI-generated answers.

Some of that work will be useful. Brands should know whether they are showing up, what the answer says, where the model appears to be pulling evidence from, and whether the information is accurate.

Adobe is clearly building for this market, with Brand Visibility, LLM Optimizer and Semrush now part of the picture. The important next step is connecting that visibility back to commercial outcomes.

For Qantas, that means not just whether the brand appears in an AI answer, but whether that journey eventually helps someone book a flight, hotel or holiday, and whether the signal flows into the systems that measure the outcome.

That is where the category gets serious.

Until then, the usual caution applies. Mentions are not revenue. Citations are not trust. Share of AI answer is not the same as customer preference.

The CMO question is not whether the brand appears in ChatGPT. The CMO question is whether AI-mediated discovery changes commercial outcomes.

Does it drive qualified traffic? Does it improve conversion? Does it shift consideration? Does it reduce acquisition cost? Does it help existing customers make better decisions? Does it surface the brand accurately when the stakes are high?

Without that link back to the business, AI search visibility risks becoming another dashboard for marketing anxiety.

Executives should expect to hear a lot more about this category. They should also ask harder questions before treating it as a strategy.

The old fundamentals get more valuable

The interesting thing about the Qantas framing is that it does not make old brand work irrelevant.

It makes it more important.

If AI tools summarise the market by drawing from many sources, then the quality of the underlying business matters more, not less. Product quality matters. Service experience matters. Pricing clarity matters. Accessibility matters. Reviews matter. Support content matters. Reputation matters.

This is not just a B2C issue.

In B2B, the same pattern applies. A buyer might use an AI tool to shortlist vendors, compare categories, draft an RFP, summarise analyst reports or understand implementation risks. But the buyer still has to trust the vendor enough to put budget, reputation and operational risk behind the decision.

The AI answer may accelerate discovery.

It does not eliminate procurement, proof, politics or accountability.

For leadership teams, that means AI search is not just a marketing issue. It touches customer experience, product, support, communications, reputation, sales enablement and risk.

The answer a model gives about your company may be a marketing moment. The reasons it gives that answer are often operating issues.

The useful stoic idea

Walters also reached for an old stoic idea, often paraphrased through the Serenity Prayer:

That is a useful way to think about AI search.

A brand cannot control exactly what every model says about it. It cannot control every forum post, review, comparison, complaint or third-party summary. It cannot force a customer to trust the first AI-generated answer.

But it can influence the evidence.

It can make its own content clearer. It can fix inaccurate information. It can improve the customer experience that generates reviews. It can reduce support friction. It can make product claims easier to verify. It can build a reputation that survives the summary layer.

That is the practical work.

Not gaming the model. Making the brand easier to trust when the model points at it.

The CMO takeaway

AI search matters. Brands should not ignore it.

But the wrong response is to treat it as SEO with a new coat of paint. The better response is to treat it as a pressure test of the brand’s wider evidence base.

If customers are asking AI tools what to buy, who to trust, which provider is reliable, or which option best fits their needs, the model is not just looking for your preferred message. It is looking for signals.

That should make executives uncomfortable in a useful way.

Because the question becomes bigger than: do we rank?

It becomes: what does the market actually know about us, and would we trust the answer if we were the customer?

AI search is not the customer.
It is the new middle layer.

The customer is still human. The decision is still about trust. And the brands that win will be the ones with the strongest evidence, not just the best prompt strategy.

Disclosure: Adobe invited me to Adobe Summit Sydney and covered travel and accommodation. Adobe had no editorial review or approval. I’m interested in practical AI stories, not vague transformation language or paid placement. Pitch one here.

Agents. Orchestration. Brand visibility. AI teammates. Personalisation at scale. Humans and machines. Customer experience flywheels.

Some of that matters. Some of it is just vendor language doing what vendor language does. But the most useful story of the day was not the most futuristic one.

It was RACQ.

Not because RACQ had the flashiest AI demo. It didn’t. RACQ was useful because its story showed what enterprise AI usually looks like once it leaves the keynote stage and has to survive inside a real organisation.

Less magic. More plumbing. More trust.

Trust comes first

RACQ is a large Queensland member organisation operating across roadside assistance, insurance, banking, batteries, solar and travel. It sits in the kind of category where customer experience is not a slogan. People turn to RACQ when something matters.

That makes its AI story more useful than a polished demo.

Tim Cochrane, RACQ’s General Manager of Marketing, Membership and Digital, kept coming back to the same point. RACQ is a brand built on trust. Whatever marketing method it uses, that trust is non-negotiable.

That is a better starting point for AI than most of the conference language.

The first question is not what the technology can do. The first question is what the brand can do without weakening the trust that makes the technology useful in the first place.

Subscribe now

The problem was operational

RACQ’s issue was not that it lacked access to clever tools. The problem was more basic. Its systems were manual, fragmented and siloed.

The organisation was trying to serve members across multiple products, but did not have the kind of unified customer understanding needed to do that well. In plain English, it was hard to know enough about the member, at the right moment, across the right part of the business.

That is not a model problem. It is an operating problem.

This is the part most AI strategy conversations skip over. They start with what the technology can now generate, automate or personalise. The better executive question is what the organisation is ready to let it touch.

If the data is fragmented, the workflows are brittle and the customer view is incomplete, adding AI does not create a smarter organisation. It just creates faster movement through a messy one.

The boring work paid back

The strongest RACQ proof point was not an agent. It was a customer data story.

Cochrane talked about the work to create a better view of the customer. That sounds dull until you hear what it changed.

Some RACQ households had been receiving multiple copies of the same glossy magazine. In the worst cases, one household could receive four copies.

That is not an AI problem. It is a customer data problem. It is also the kind of problem that tells you whether an organisation is actually ready for more advanced automation.

If you cannot work out that one household should not receive four copies of the same magazine, you probably should not be talking too confidently about AI agents managing the customer journey.

The payoff was not just less waste. RACQ could also better identify people who had sought an insurance quote but did not transact, then re-engage them with more relevant timing and context.

Cochrane said that helped drive a significant increase in conversion. He also said RACQ had planned for payback over roughly two and a half years, but achieved it in about nine months from go-live.

That is the kind of AI-adjacent story executives should care about.

Not because the number is enormous. Because the mechanism is understandable. A customer shows intent, the business recognises that intent, the business responds with better timing and relevance, and conversion improves.

That is not science fiction. It is better execution.

The AI ad disclosure matters

There was another RACQ detail that says a lot about where AI marketing is heading.

RACQ has created video advertising using AI-generated visuals, built around Australian animal characters Sam and Reg. The writing, production and voice work still involved humans, but the visuals were generated using AI.

The important detail is that RACQ clearly discloses this in the ad: “Visuals generated using AI.”

That may sound like a small compliance note. I do not think it is.

Before the generative AI era, RACQ had used CGI. That is also computer-generated, of course. But CGI and AI-generated visuals now carry different expectations for audiences, regulators and brand teams.

CGI usually implies a controlled production process involving artists, animators, rendering, compositing and approvals. AI-generated visuals introduce a different trust question because the synthetic layer is more automated, more scalable and less intuitively understood by the audience.

That does not make AI-generated advertising wrong. It means the brand has to think harder about disclosure.

For RACQ, that matters because trust is not a decorative brand value. It is central to the business. If members believe the organisation is using synthetic media without being straight with them, the production saving may not be worth the reputational cost.

That is the practical AI question for brands. Not just whether AI can make something cheaper or faster, but whether it can be used in a way that still feels honest.

The agent comes later

Adobe’s event framing was heavy on agents and orchestration. RACQ’s story points to a more useful sequence.

First, fix the customer data and workflow foundations. Then improve the speed and relevance of engagement. Then layer in more advanced AI and automation where the organisation has enough trust, governance and context to use it safely.

A lot of executive AI work still starts in the wrong place. It asks what the agent can do, which tasks it can complete, and which workflows it can automate.

Those are useful questions, but they are not the first ones.

The first questions are more basic. What does the organisation know about the customer? Which systems hold that knowledge? Which teams act on it? Which approvals slow it down? Which customer moments are commercially meaningful? Which data is trusted enough to drive action?

Only then does the agent matter.

The hidden trade-off

There is also a harder executive question underneath the RACQ story.

RACQ has entered a five-year strategic partnership with Adobe and Deloitte Digital. The official announcement talks about a Lighthouse model, early access to Adobe AI capabilities, and a governance and learning framework. That is the formal version. The practical version is simpler: RACQ is making a serious long-term bet on a major enterprise stack and the services required to make it work.

That may be the right decision. Large organisations often need serious platforms and serious partners to make change stick.

But CEOs should not pretend the trade-off disappears because the word AI is attached.

When AI becomes embedded in customer data, content workflows, member engagement, advertising, approvals and decisioning, the vendor is no longer just selling software. It is helping shape how the organisation operates.

That can create leverage. It can also create lock-in.

The right question is not whether that is good or bad in the abstract. The right question is whether the organisation understands what it is standardising around, what it can still change later, and which capabilities it must own internally.

The lesson for CEOs

RACQ’s story cuts through the easiest version of the AI narrative.

Enterprise AI value does not begin with a chatbot. It does not begin with an agent. It does not begin with a demo that turns a prompt into a campaign.

It begins when an organisation can see enough of the customer, trust enough of the data, coordinate enough of the workflow and measure enough of the result to act differently.

It also begins with a clear view of what the brand will not compromise. For RACQ, that is trust. The method of marketing delivery can change, but the trust obligation cannot.

The companies that get value from AI will not be the ones with the most experiments. They will be the ones that turn AI into better operating discipline.

RACQ’s case is still early. The next test is whether the five-year partnership produces durable member value, not just better marketing execution or a cleaner technology stack.

But as a signal, it is useful.

The boring work is not the prelude to enterprise AI. It is enterprise AI.

Share

Disclosure: Adobe invited me to Adobe Summit Sydney and covered travel and accommodation. Adobe had no editorial review or approval. I’m interested in practical AI stories, not vague transformation language or paid placement. Pitch your story here.

In the latest season of The Capture, a powerful AI system called Simon sits at the centre of critical government decision-making. Simon analyses information, evaluates options and recommends actions faster than any human team could. At first, the people around it remain firmly in control. They review its recommendations, challenge its conclusions and make the final decisions themselves.

Over time, however, something subtle happens.

Simon proves itself useful often enough that trust begins to replace scrutiny. Nobody formally hands over authority, but fewer and fewer decisions are made without Simon’s involvement. The system becomes so embedded in the organisation that its absence becomes almost unthinkable.

While the story is fictional, the leadership challenge is not.

Across every industry, organisations are embedding AI into hiring, forecasting, pricing, customer service, risk assessment and strategic planning. Most discussions about AI focus on capability: what the technology can do, how much productivity it can unlock and where costs can be removed. These are important questions, but they are not the only questions CEOs should be asking.

An equally important question is what happens to organisational capability when AI begins performing work that previously developed human judgement.

Subscribe now

From Leadership Dependency To System Dependency

For decades, leadership teams have worked hard to eliminate dependency on key individuals. Many of us have spent years building management capability, developing future leaders, introducing succession plans and distributing decision-making authority throughout our organisations. The logic was simple. Organisations scale when knowledge, judgement and ownership become distributed. They struggle when expertise remains concentrated in a handful of people.

AI introduces a very different path to scale.

Rather than creating more capable decision-makers, organisations can increasingly rely on systems that provide expertise, analysis and recommendations on demand. Need a forecast? Ask the model. Need market analysis? Ask the model. Need a recommendation? Ask the model. In many cases the answer is faster, cheaper and often better than the one available through traditional processes.

The productivity gains are real.

The risk is that capability may no longer be growing at the same rate as dependency.

This distinction matters because capability and dependency are not opposites. It is entirely possible for an organisation to become more productive while simultaneously becoming more dependent. We have seen this pattern before in other forms of technology adoption. Systems become more sophisticated, outputs become more reliable and organisations become increasingly confident in their use. Eventually the technology becomes embedded so deeply into daily operations that few people remember how to perform the underlying activity without it.

When Efficiency Replaces Capability

Recruitment provides a useful example.

Many organisations now use AI-assisted tools to screen, rank and prioritise candidates before a hiring manager ever reviews an application. Recent legal action involving Workday has focused attention on the role AI plays in hiring decisions and whether organisations fully understand the consequences of algorithmic recommendations.

What interests me more is the capability question.

How many hiring managers today are evaluating candidates, and how many are evaluating AI-generated shortlists?

Those activities may appear similar, but they develop very different skills. Over time, fewer managers gain experience identifying talent independently because the machine performs the initial analysis. The process becomes more efficient, but efficiency and capability are not always the same thing.

A similar pattern is emerging in knowledge work. Reports emerged that a major KPMG AI-related report contained fabricated references and examples generated through AI-assisted research processes. The issue was not that AI produced inaccurate content. Every executive who uses these tools understands that errors occur. The more interesting issue was that the content appeared sufficiently credible to pass through professional review processes.

The failure was not technological.

It was organisational.

People reviewing the material appeared to place greater trust in the output than they would have if the same work had been produced through traditional means. Authority had subtly shifted from the reviewer to the system.

Neither example represents a failure of AI.

Both examples represent a failure to maintain human capability alongside AI capability.

The Aviation Warning

Other industries have wrestled with similar challenges for years.

Aviation is often cited as one of the most successful examples of automation improving safety and consistency. Modern aircraft are extraordinarily sophisticated, and automation has contributed significantly to reductions in operational risk. Yet aviation experts have also spent decades discussing automation dependency and skill degradation. The concern is not that automation is harmful. The concern is that skills deteriorate when they are rarely exercised.

Pilots who spend less time manually flying aircraft may become less proficient at handling unusual situations when automation is unavailable. Airlines have spent years redesigning training programs to ensure pilots maintain critical skills despite increasing levels of automation.

The same question is beginning to emerge across knowledge work.

If AI performs the analysis, who develops analytical capability?
If AI produces the recommendation, who develops judgement?
If AI writes the first draft, who develops communication capability?

These questions become increasingly important as AI adoption moves from experimentation into core operational workflows.

The Metric Nobody Is Measuring

This is where the discussion becomes particularly relevant for CEOs.

Most organisations currently measure AI success through productivity improvements, cycle-time reductions, automation rates and cost savings. Those metrics are sensible and necessary. They help justify investment and demonstrate operational impact.

What they do not reveal is whether organisational capability is becoming stronger or weaker as AI adoption expands.

A business may improve efficiency dramatically while simultaneously reducing the number of people capable of making high-quality decisions without technological assistance. Over time, that can create a new concentration risk. Historically, organisations worried about dependency on key executives. Today they may need to worry about dependency on key systems.

The risk is not that AI becomes intelligent enough to replace leadership.

The risk is that leaders and teams become so accustomed to AI-supported decision-making that they gradually lose confidence in operating without it.

Capability migrates from people into platforms, often so gradually that nobody notices it happening.

The AI Dependency Audit

A useful exercise for executive teams is remarkably simple.

Imagine every AI system in your organisation became unavailable for the next thirty days. No copilots, no AI-generated reports, no forecasting assistants and no automated analytical tools.

Now ask four questions:

• Which decisions would slow down?

• Which teams would struggle most?

• Which leaders would become bottlenecks?

• Which capabilities would suddenly become scarce?

The answers are often revealing.

Every dependency exposed by that exercise represents a capability that may no longer exist where you think it does. Just as leadership teams conduct succession planning for key executives, they may soon need to understand their succession risk for critical AI-supported capabilities.

Final Strategic Truth

For decades, organisations focused on reducing dependency on key individuals. That was the right challenge for the era.

The AI era introduces a new one.

How do we capture the benefits of machine intelligence without allowing human judgement to quietly erode?

Because the objective was never to remove people from decision-making.

The objective was to create more people capable of making good decisions.

Share

The CEO Question

If every AI system in your organisation disappeared tomorrow, would your people still know how to think?

The answer may reveal whether you are building capability at scale, or quietly outsourcing it.

Yet many CEOs experience the opposite. More decisions seem to require executive involvement. More meetings appear on the calendar. More issues are escalated for review. Despite having more leaders than ever before, the organisation becomes increasingly dependent on a small number of people to keep things moving.

The result is familiar. Projects slow down waiting for approvals. Leadership teams become overloaded. Talented people stop taking ownership because they know major decisions will eventually be referred upwards anyway.

What appears to be a capability problem is often something else entirely.

It is a dependency problem.

The Scaling Trap

This pattern is surprisingly common because it is often created by the same leadership behaviours that drove success in the first place.

Most CEOs reach leadership positions because they are good at solving problems. They make decisions quickly. They provide clarity in uncertain situations. They have experience that others rely on. When organisations are small, these qualities create momentum and accelerate growth.

As organisations become larger, however, the economics of leadership begin to change.

The number of decisions grows faster than executive bandwidth. New markets create new complexity. More customers create more edge cases. More employees create more management challenges. The leadership model that worked at 50 people becomes difficult to sustain at 500.

What once accelerated the organisation can gradually begin to constrain it.

Subscribe now

The Hidden Cost Of Having The Answers

Many executives assume their value comes from having the best answers in the room.

It is an understandable assumption. Leaders are rewarded for expertise. Boards expect judgement. Teams look to executives for guidance. Over time, leaders become accustomed to being the person people turn to when something important needs to be resolved.

The problem is that organisations learn from behaviour.

When leaders consistently provide solutions, people bring them problems. When leaders make the key decisions, people wait for direction. When leaders intervene quickly, ownership gradually moves upwards rather than outwards.

None of this happens intentionally.

In fact, many leadership teams create dependency while trying to be helpful.

The Framework Behind The Idea

Leadership researcher Liz Wiseman spent years studying why some leaders seemed to expand the capability of the people around them while others unintentionally reduced it.

Her conclusion was simple.

Some leaders act as Multipliers. They create environments where people contribute more, think more deeply and take greater ownership of outcomes. Other leaders act as Diminishers. They become the primary source of answers, decisions and direction, causing capability to concentrate around them rather than spread throughout the organisation.

The distinction is important because it reframes the role of leadership.

The objective is not simply to make better decisions personally. The objective is to increase the number of people throughout the organisation who are capable of making good decisions without requiring executive intervention.

That shift becomes increasingly important as organisations scale.

Why This Matters More In The AI Era

For decades, organisations operated on the assumption that expertise was scarce. Information was difficult to access. Analysis was expensive. Experience was concentrated among a relatively small group of senior leaders.

Today that assumption is becoming less true.

AI can generate options, analyse information, challenge assumptions and provide expertise on demand. While judgement remains valuable, access to knowledge is no longer the constraint it once was.

The challenge facing many organisations is not finding better answers.

The challenge is creating enough people who are confident and capable enough to act on them.

As AI continues to democratise expertise, the leaders who create the greatest value will not necessarily be those with the most answers. They will be the leaders who create organisations capable of making good decisions at scale.

The Leadership Dependency Test

One of the simplest ways to identify organisational dependency is to imagine that you are unavailable for the next 30 days.

No meetings. No email. No phone calls. No messages.

Now ask yourself four questions.

What decisions would stop?
What projects would slow down?
What customer issues would escalate?
What meetings would be cancelled?

The answers are often revealing. Every stalled decision, delayed project or unresolved issue highlights an area where the organisation remains dependent on executive involvement.

Some dependencies are entirely appropriate. Many are not.

The exercise helps leaders identify where capability has failed to spread and where ownership still sits too close to the top of the organisation.

Leadership Dependency Audit Prompt

Executive Meeting Review Prompt

90-Day Multipliers Plan Prompt

The Real Value Of Multipliers

Most leadership development focuses on improving the capability of individual leaders. Multipliers focuses on improving the capability of the entire organisation.

That distinction matters because organisations do not scale through individual performance. They scale when decision-making, ownership and problem-solving become distributed across larger numbers of capable people.

The most effective leaders are rarely those who make every important decision themselves. They are the leaders who create conditions where good decisions can be made without them.

Liz Wiseman’s research provides a useful reminder that leadership is not simply about what leaders achieve personally. It is about what they enable others to achieve collectively.

As organisations grow, that difference becomes increasingly important.

The Question Every Leadership Team Should Answer

If your executive team disappeared for the next 30 days, what would stop?

The answer may reveal whether your organisation is building capability at scale—or concentrating it at the top.

Share

If you found this useful, share it with an executive who is trying to scale an organisation without becoming its bottleneck.

The largest IPO in history generated exactly the debate you would expect.

Investors argued about valuation. Analysts debated Starlink. Journalists dissected Elon Musk’s influence. Everyone seemed to have an opinion on launch economics, subscriber growth, competitive positioning, and whether SpaceX deserved its place among the most valuable companies in the world.

After spending the weekend reading the prospectus, I found myself focused on a completely different question.

Not whether the company is worth its valuation.
Not whether Starship succeeds.
Not whether Starlink becomes the dominant communications network on Earth.

The question that kept coming back to me was much simpler:

Because the entire filing reads differently once you start looking for constraints instead of opportunities.

Subscribe now

The Number That Made Me Stop

Most discussions about SpaceX begin with the company’s valuation. Mine started with a number buried in the cash flow statement.

In 2025, SpaceX generated $18.7 billion in revenue. It also lost nearly $5 billion. At the same time, the company spent almost $20 billion on investing activities. Those numbers are unusual on their own. Together, they tell a story. A mature company with nearly $19 billion in revenue is not normally expected to lose billions of dollars while simultaneously investing at extraordinary scale.

Most public companies lose money because something has gone wrong. SpaceX appears to be losing money because management believes something is about to happen.

That distinction matters because it changes how you interpret almost everything else in the filing.

A company optimising for current profitability allocates capital very differently from a company attempting to prepare for future constraints. The prospectus suggests SpaceX belongs firmly in the second category.

Most Companies Forecast Demand

Traditional corporate planning revolves around demand. Executives build forecasts, estimate market size, monitor customer behavior, and invest when demand becomes visible. New factories are built because orders are growing. New data centres are commissioned because utilisation is rising. New employees are hired because the sales pipeline justifies expansion.

This is rational behavior. It is also how most companies are taught to operate.

What struck me while reading the SpaceX filing is how often the company appears to be working from a different playbook. Again and again, management describes investments that are being made long before the corresponding market fully exists. Starship is being built before a meaningful lunar economy exists. Orbital AI compute is being discussed before most investors believe it is commercially viable. Terafab is being positioned as a future source of chip capacity before shortages become critical. AI infrastructure is being deployed at extraordinary scale before the economics are fully proven.

Viewed individually, these projects can appear disconnected. Viewed together, they reveal a consistent pattern.

SpaceX appears less interested in forecasting demand than in forecasting constraints.

The First Constraint: Launch

Most investors think of Starship as a rocket.

The prospectus does not.

Throughout the filing, Starship is described as the foundation for future growth across multiple businesses. Management argues that Starship could eventually reduce the cost of reaching orbit by more than 99% relative to historical launch costs, creating what it calls a scalable path toward future infrastructure, including orbital AI compute.

That is a very unusual way to think about a launch vehicle.

Most aerospace companies sell launches. SpaceX appears to be investing in the future supply curve of orbit itself. The company repeatedly links launch cadence, satellite deployment, connectivity expansion, and future AI infrastructure. In one risk factor, delays to Starship are described as a threat not only to future satellites and communications services, but also to orbital AI compute.

In other words, launch is not being presented as a product.

It is being presented as infrastructure.

The Second Constraint: Compute

The AI section of the prospectus contains one of the most revealing financial disclosures in the entire document.

The AI segment’s Adjusted EBITDA deteriorated from a positive $347 million in 2024 to a loss of $1.2 billion in 2025. Management attributes the decline largely to cloud computing costs, data centre infrastructure, facilities, and employee expenses. More importantly, the company explicitly states that the AI segment is being driven by a strategy to “rapidly and cost-effectively scale compute infrastructure.”

This is not how software companies typically talk.

Most software companies talk about users, engagement, products, and monetisation. SpaceX talks about compute capacity. It highlights gigawatt-scale training clusters, massive data centres, deployment speed, power infrastructure, and construction economics. The company boasts that it brought its first COLOSSUS cluster online in 122 days and a second cluster online in just 91 days, dramatically faster than industry norms.

The message is clear.

Management appears to believe that compute capacity itself is becoming the strategic asset.

The Third Constraint: Chips

Then there is Terafab.

If you look at Terafab in isolation, it feels strange. Why would a company known for rockets and satellites be planning one of the world’s largest chip manufacturing facilities?

The filing provides the answer.

SpaceX repeatedly argues that future AI growth depends on access to compute hardware. Terafab is described as a way to extend control to what management calls the “foundational chip layer.” The company intends to deepen vertical integration by moving closer to processor design, fabrication, packaging, and manufacturing. It explicitly frames this effort as a competitive advantage in what it calls the race to scale AI infrastructure.

Once again, the pattern emerges.

The company is investing not where demand exists today, but where it believes future bottlenecks will appear.

The Fourth Constraint: Energy

This is where the filing becomes genuinely interesting.

Most AI discussions assume that future growth requires more models, more data centres, and more GPUs. SpaceX agrees that more compute will be needed. It disagrees on where that compute should ultimately live.

The prospectus repeatedly argues that Earth’s finite resources will eventually struggle to support future computational demand. Management states that sustaining future AI growth will require space-based infrastructure powered by the Sun. The company describes orbital compute as a long-term response to power, cooling, and infrastructure constraints that it believes will eventually limit terrestrial systems.

Whether you agree with this conclusion is almost beside the point.

The important observation is that management clearly believes it.

And if management believes it, then much of the company’s current investment strategy starts to make more sense.

Share

Suddenly Everything Fits Together

At this point, Starship, Starlink, COLOSSUS, Terafab, and orbital compute stop looking like separate initiatives.

They begin to look like different responses to the same anticipated problem.

Starship addresses launch constraints.
Starlink addresses connectivity constraints.
COLOSSUS addresses compute constraints.
Terafab addresses chip constraints.
Orbital compute addresses energy constraints.

Each project sits at a different layer of the stack. Each project targets a different bottleneck. Each project requires significant capital long before corresponding demand is obvious.

That does not mean management is right.

But it does mean the strategy is more coherent than it first appears.

The Strategic Lesson Hidden In The Filing

The lesson here is not that every company should build rockets, data centres, or chip factories.

The lesson is that the most consequential strategic decisions are often made years before a bottleneck becomes visible to everyone else.

Most executives spend their time asking how large a market might become. That is a necessary question. SpaceX appears to be asking a different one.

What breaks first if the market becomes enormous?

That is a much harder question to answer. It is also often the more valuable one.

History is full of examples. Railroads were not the Industrial Revolution, but they controlled a critical constraint. Telecommunications networks were not the Internet, but they enabled it. Cloud providers were not the software industry, but they became some of its most powerful participants.

The prospectus suggests SpaceX believes AI will follow a similar pattern.

The Question Investors Should Be Asking

The market will spend years debating whether SpaceX deserves its valuation. That debate is inevitable, and reasonable people will disagree.

The more interesting question is whether investors have correctly identified what the company is actually building.

The prospectus reads less like a company optimising for today’s markets and more like a company preparing for tomorrow’s constraints. Again and again, management points to launch capacity, compute capacity, chip capacity, and energy capacity as the foundations upon which future growth will depend.

The market thinks SpaceX went public as a rocket company.

After reading the filing, I think SpaceX is trying to build capacity ahead of physics.

Current revenue. Current customers. Current headcount. Current constraints. Current market conditions.

Then they ask a seemingly reasonable question:

What should we do next?

The problem is that starting from the present often produces incremental thinking.

If your planning process begins with today’s constraints, today’s resources and today’s assumptions, it becomes difficult to imagine a fundamentally different future. Strategic discussions drift toward optimisation rather than transformation.

The result is familiar.

A larger sales team.
A few new product features.
Some operational improvements.
A slightly better version of the business that already exists.

Useful, but rarely game-changing.

The most ambitious organisations take a different approach.

They start with the destination.
Then they work backwards.

The Principle Behind The Framework

Stephen Covey described this idea decades ago in The 7 Habits of Highly Effective People.

The concept sounds simple.

Yet organisations struggle to apply it consistently because they immediately jump from vision to execution. They discuss where they want to go and then rush into project plans, budgets and quarterly objectives.

What is often missing is the bridge between the future and the present.

Amazon built an entire management process around that bridge.

They call it Working Backwards.

Amazon’s Working Backwards Framework

Amazon’s approach became famous because it reverses the sequence most companies use.

Rather than starting with resources and asking what can be built, teams start with the customer outcome and ask what must be true to achieve it.

Before launching a product, teams write the future press release.
Before discussing implementation, they describe success.
Before debating priorities, they define the outcome customers will experience.

Only after everyone agrees on the destination do they begin planning the journey.

The process forces a different conversation.

Instead of asking:

“What should we build?”

Teams ask:

“If we were wildly successful, what would customers be saying?”

That shift sounds subtle.

In practice, it changes everything.

Subscribe now

Why Strategic Planning Fails

Many strategic planning exercises become exercises in resource allocation.

Which initiatives should we fund?
Which markets should we prioritise?
Which projects should we accelerate?

Those are important questions.

But they assume leadership has already agreed on what success looks like.

Often they haven’t.

The absence of a clearly defined future creates a hidden problem. Every executive carries a different mental model of where the organisation is heading.

The CEO imagines market leadership.
The CFO imagines margin expansion.
The Head of Product imagines innovation leadership.
The COO imagines operational excellence.

None of these goals are wrong.

But they are not necessarily the same.

As a result, organisations become highly productive while remaining partially misaligned.

People work hard.
Resources get allocated.
Projects get delivered.

Yet progress feels slower than it should because the organisation is pursuing multiple versions of success simultaneously.

Working Backwards solves this problem by forcing leaders to describe the destination in enough detail that everyone can see it.

The Future Press Release

The first step is deceptively simple.

Imagine it is 36 months from now.
Your company has exceeded expectations.
Customers love the product.
The business is thriving.
Competitors are paying attention.
Industry analysts are writing about your success.

Now write the press release announcing that outcome.

Not the strategy.
Not the roadmap.
The result.

This exercise forces leaders to articulate success in concrete language rather than vague aspirations.

One of the fastest ways to do this is with AI.

Future Press Release Prompt

The FAQ Test

Amazon doesn’t stop with the press release.

The next step is often more valuable.

Teams write the FAQ.

The reason is simple.

A future vision is easy to agree with.
The details are where strategy becomes real.

The FAQ forces leaders to confront difficult questions before committing resources.

What assumptions must be true?
What obstacles might derail us?
What would investors question?
What would customers challenge?
What capabilities are missing today?

Many strategic plans fail because organisations become emotionally attached to the vision without rigorously testing the assumptions underneath it.

The FAQ exposes those assumptions early.

Future FAQ Prompt

The Reverse Roadmap

Many organisations naturally think forwards.

Today leads to tomorrow.
Tomorrow leads to next quarter.
Next quarter leads to next year.

Working Backwards flips the process.

Once the destination is clear and the assumptions have been tested, leaders can begin mapping the path in reverse.

Starting from the future often reveals priorities that are invisible when planning from the present.

Capabilities that seemed optional become essential.
Projects that looked urgent become distractions.
Investments that felt risky become obvious.

The roadmap becomes less about activities and more about milestones that must exist for the future state to become possible.

Reverse Roadmap Prompt

The Real Value Of Working Backwards

Most leaders assume strategy is about deciding what to do.

The best leaders understand that strategy is first about deciding what success looks like.

Once that becomes clear, many decisions become easier.

Opportunities can be evaluated against the destination.
Investments can be prioritised.
Trade-offs become obvious.
Alignment improves.
Focus improves.
Execution improves.
The future becomes a filter for the present.

This is the deeper lesson behind Covey’s principle.

Beginning with the end in mind is not about vision boards or motivational slogans.

It is about creating enough clarity about the future that today’s decisions become easier to make.

Amazon simply turned that idea into a repeatable management system.

The Question Every Leadership Team Should Answer

If your company became extraordinarily successful over the next 36 months, what would the press release say?

If your leadership team cannot answer that question clearly, there is a good chance you are planning from the present instead of leading from the future.

Share

If you found this useful, share it with a CEO or executive who is in the middle of a strategic planning cycle.

Every few months another headline appears warning that AI is consuming too much electricity, too much water, or too many resources.

The implication is usually the same.

AI is bad for the planet.

There is truth in that claim.

Training and operating large AI models requires enormous computing infrastructure. Data centres consume electricity. Cooling systems use water. New facilities require land, construction materials and transmission infrastructure.

According to the International Energy Agency, data centre electricity demand is expected to grow significantly over the coming decade, driven largely by AI workloads.

The environmental costs are real.

Ignoring them helps nobody.

But there is another side to the story.

The Headlines Aren’t Wrong

Most public discussion focuses on total consumption.

More data centres.
More power.
More water.

What gets overlooked is that the technology itself is improving at an extraordinary pace.

Every generation of AI hardware performs more work with less energy. Every generation of AI software becomes more efficient. Every generation of AI models delivers more capability from the same infrastructure.

This is not unique to AI.

It is the pattern that has defined computing for decades.

The environmental cost per unit of intelligence is falling.

That distinction matters.

Subscribe now

But They Miss Something Important

Cars became more fuel efficient.
Computers became more energy efficient.
Air travel became more fuel efficient.

Yet society ultimately used more of each because they became more useful, more accessible and more affordable.

AI is likely to follow the same path.

Even if individual AI tasks become dramatically more efficient, total usage may continue to grow because the technology becomes increasingly valuable.

Which means two things can be true at the same time.

AI is becoming greener.
AI is consuming more resources.

We’re Asking The Wrong Question

The debate around AI often starts with the wrong question.

“Is AI sustainable?”

Nobody asks whether hospitals are sustainable.
Nobody asks whether universities are sustainable.
Nobody asks whether the internet is sustainable.

All of them consume resources.

The real question is whether the value they create justifies the resources they consume.

AI should be evaluated the same way.

Not Every Outcome Is Worth The Cost

Using AI to generate endless streams of low-value content is very different from using AI to improve logistics, optimise energy systems, reduce waste, accelerate scientific discovery or increase productivity.

The technology is the same.
The outcome is not.

Research highlighted by the International Energy Agency suggests AI could improve electricity networks, building efficiency, industrial processes and transportation systems.

Researchers publishing in Nature Communications also found significant opportunities for AI to improve energy efficiency in buildings, one of the world’s largest sources of emissions.

If AI helps reduce waste across the broader economy, then the comparison is not between AI emissions and zero.

It is between AI emissions and the emissions that never occur because better decisions were made.

The Market Is Doing Its Job

My instinct has always been that markets eventually determine what survives.

Every AI provider must pay for chips, electricity, buildings, cooling systems and talent.

If those inputs become expensive, providers have a powerful incentive to reduce consumption and improve efficiency.

Competition rewards those who can deliver more intelligence with fewer resources.

That is one reason computing has become relentlessly more efficient over time.

Markets are remarkably good at squeezing waste out of systems.

But they are not perfect.

Carbon emissions, water scarcity and local environmental impacts are not always reflected in the price of a transaction.

That does not invalidate markets.

It simply means leaders should understand their limitations.

Share

Someone Still Has To Decide What’s Worth Doing

For CEOs, CFOs and boards, the environmental debate around AI is not really about AI.

It is about stewardship.

Every organisation consumes resources in pursuit of outcomes.

Capital.
Energy.
Materials.
Talent.
Time.

AI is simply another input.

The challenge is ensuring the value created exceeds the resources consumed.

The environmental cost of AI is real.

The value of AI can also be real.

Leadership requires understanding both sides of the ledger.

What We Do With It Matters

AI is neither an environmental catastrophe nor a free lunch.

It is infrastructure.

Like every major technology before it, it consumes resources while simultaneously becoming more efficient.

The environmental debate should not be about whether AI exists.

It should be about whether we are using increasingly affordable intelligence to create outcomes that are genuinely worth the cost.

A transformation program falls behind schedule and leadership asks for a more detailed plan. A growth initiative misses expectations and another strategy review is commissioned. An AI rollout generates mixed results and governance expands. A new market opportunity emerges and the executive team spends months refining forecasts before making a move. In each case, intelligent people are trying to reduce risk. In each case, they may be making the situation worse.

The instinct is understandable. Executives are rewarded for being right. Boards reward confidence, investors reward predictability and organisations reward leaders who appear to have answers. Over time, this creates a powerful bias toward certainty. When uncertainty increases, leadership teams naturally respond by seeking more information, more analysis and more validation. Sometimes that is exactly the right response. The problem is that many of the most important challenges facing organisations today do not become clearer through analysis alone.

This distinction sits at the heart of a surprisingly common executive mistake. Leaders often assume every challenge can be solved using the same decision-making process. Gather the facts. Analyse the options. Build the business case. Execute the plan. That sequence works extremely well in some situations. It performs remarkably poorly in others. The issue is not the quality of the leaders or the quality of the analysis. The issue is that different categories of problems require fundamentally different approaches.

The certainty trap

Most executives spend their careers solving problems where expertise and analysis create a measurable advantage. A CFO evaluating an acquisition can improve the quality of a decision through deeper financial analysis. A COO redesigning a supply chain can model scenarios and optimise trade-offs. A CIO selecting a technology platform can compare vendors, assess risks and make a rational recommendation. These are difficult decisions, but they share a common characteristic. The answer exists. The organisation may not know it yet, but the answer can be discovered through expertise, investigation and analysis.

Success in these environments creates a powerful habit. Leaders become accustomed to the idea that uncertainty is simply a temporary lack of information. If the business gathers enough data, speaks to enough experts and performs enough analysis, the right answer will eventually emerge. The habit is reinforced because it works so often. The challenge is that many of today’s most important executive questions do not behave this way.

Culture change does not behave this way. Product innovation does not behave this way. Building a new market category does not behave this way. Responding to disruptive technology does not behave this way. In these situations, the future is not waiting to be discovered. It is being created through the actions of customers, competitors, regulators and the organisation itself. The answer does not exist yet. It emerges over time.

This is where organisations become trapped. Faced with uncertainty, they reach for the tools that have worked in the past. More planning. More governance. More analysis. More certainty. The result is that leadership teams become increasingly sophisticated at discussing the future while becoming progressively slower at influencing it. They mistake activity for learning and preparation for progress.

The framework that explains the mistake

Dave Snowden developed the Cynefin Framework after observing that organisations repeatedly struggled not because they lacked intelligence, but because they misclassified the problems they faced. Leaders would apply a management approach that was successful in one environment and assume it would work equally well in another. The problem was not execution. The problem was diagnosis.

The framework itself is deceptively simple. Some challenges are clear. The answer is known, best practices exist and consistency creates value. Some challenges are complicated. The answer exists but requires expertise to discover. Some challenges are complex. The answer emerges through experimentation and learning. Some challenges are chaotic. The immediate priority is stabilisation before deeper understanding becomes possible.

The power of the framework is not in the categories themselves. The power comes from recognising that different categories require different forms of leadership. A complicated problem benefits from analysis. A complex problem benefits from experimentation. A chaotic problem benefits from decisive action. The mistake most organisations make is treating all three as though they are merely complicated.

When that happens, transformation programs become trapped in governance structures designed for operational projects. Innovation initiatives become trapped in planning cycles designed for capital investments. Crises become trapped in committee discussions. The organisation applies the wrong operating system because it has misunderstood the nature of the challenge.

The executive test

Many strategy discussions begin with a familiar question: What should we do? The problem is that this question often arrives too early. Before discussing solutions, leadership teams should first ask what kind of problem they are solving. Is the answer already known? Can it be discovered through expertise? Must it emerge through experimentation? Or is the situation so unstable that immediate action matters more than analysis?

Very few organisations ask these questions explicitly. As a result, they spend months debating solutions to problems they have not correctly diagnosed. They argue about forecasts before understanding whether forecasting is even possible. They debate implementation plans before determining whether the answer needs to be discovered through experimentation. They optimise the wrong decision process because they never classified the challenge in the first place.

Use this prompt to pressure-test your assumptions before committing significant time, capital or attention.

Prompt: Problem Classification Audit

Subscribe now

The Executive Decision Standard

The practical application of Cynefin is to stop treating every decision as though it belongs in the same category. Most organisations have a single decision-making culture. The same governance process is applied to innovation programs and compliance initiatives. The same reporting cadence is applied to exploratory experiments and operational projects. The same demand for certainty appears regardless of whether certainty is actually available. This creates organisational drag because the management system no longer matches reality.

A useful leadership discipline is to distinguish between discoverable problems and learnable problems. Discoverable problems contain answers that can be uncovered through expertise, analysis and investigation. Learnable problems contain answers that only emerge through experimentation and feedback. Many organisations instinctively treat both categories the same way. They ask for more evidence when they should be generating evidence. They ask for better forecasts when they should be testing assumptions. They ask for confidence when they should be seeking learning.

This distinction matters because complex environments reward a completely different capability. Success does not come from having the most accurate prediction. Success comes from reducing uncertainty faster than competitors. The organisation that learns faster gains an advantage over the organisation that plans longer. That is why some companies appear to move confidently into new markets while others spend years analysing the opportunity. The difference is often not intelligence. It is learning velocity.

The learning organisation

Many executives describe their companies as learning organisations, but few define what that means operationally. In practice, a learning organisation is one that can systematically identify assumptions, test them quickly and adapt based on evidence. Its objective is not to avoid mistakes. Its objective is to discover reality faster than competitors. That capability becomes especially valuable in complex environments where analysis alone cannot reveal the answer.

The challenge for most leadership teams is that assumptions often remain hidden. They sit inside business cases, strategy documents and executive presentations without being explicitly identified. As a result, debates occur around conclusions rather than the assumptions that produced them. Progress accelerates when leadership shifts its focus from defending plans to testing assumptions.

Use this prompt to build a learning process around a complex challenge.

Prompt: Experiment Design Workshop

The board conversation

One of the most useful applications of Cynefin is at board level. Boards are naturally skilled at challenging analysis. Directors ask whether assumptions are realistic, whether financial projections are credible and whether risks have been appropriately considered. Those questions are essential when management is dealing with a complicated problem. They become less useful when management is dealing with a complex one.

When the challenge is complex, the board should spend less time asking whether management has the answer and more time asking whether management is learning quickly enough to discover it. That subtle shift changes the entire conversation. Instead of debating forecasts, the discussion focuses on experiments. Instead of seeking certainty, the board seeks evidence. Instead of asking whether management is correct, directors ask whether management is learning.

Use this prompt before your next board discussion involving strategic uncertainty.

Prompt: Board Challenge Simulator

The executive takeaway

Most leadership teams believe their biggest challenge is making better decisions. More often, the challenge is understanding what kind of decision they are making. A board that demands certainty from a complex problem will slow the organisation down. An executive team that treats a crisis like a strategy workshop will make things worse. A leadership group that mistakes experimentation for analysis will struggle to learn.

The first responsibility of leadership is not choosing the answer. It is correctly diagnosing the nature of the problem. Because once that becomes clear, the right management approach usually becomes obvious. And until it does, even smart organisations can spend years asking the wrong question.

AI is not just software, it is power

It decides what people see, what gets approved, who gets hired, who gets served, who gets watched, who gets replaced, and who gets ignored.

Pope Leo XIV’s core warning is crisp:

That should make every CEO uncomfortable.

Because most companies are not building AI around values.

They are building it around speed, cost, scale, and margin.

Those are fine goals. But they are not enough.

If AI is trained to cut cost, it may cut corners.
If it is trained to maximise engagement, it may feed addiction.
If it is trained to optimise labour, it may treat people as waste.
If it is trained to personalise everything, it may quietly manipulate everyone.

That is the Pope’s point.

Not “AI bad.”

More like: AI makes your hidden operating system visible.

If your culture is extractive, AI can extract faster.
If your culture is sloppy, AI can scale the slop.
If your culture is paranoid, AI can become surveillance.
If your culture is humane, AI can help.

The warning is simple: do not build Babel.

Babel is the biblical story of people building a tower to prove their power. Big, clever, impressive, and doomed.

It lost the plot because the tower became the mission: power, pride, and control replaced purpose.

That is the risk with bad AI strategy.

The dashboards improve. The system gets faster. The automation spreads. But no one can explain who owns the outcome, where the judgment sits, or why customers and workers should trust it.

That is not transformation.
That is Babel in business dress.

CEOs love scale. Fair enough.

But AI does not just help you scale.
It helps you scale what you already are.

The Pope is asking: scale of what?

Scale help, or scale harm?
Scale judgment, or scale noise?
Scale trust, or scale control?
Scale human ability, or scale human replacement?

That is the whole game.

Here is my take of the papal letter.

Subscribe now

AI needs owners

Not committees. Owners.

If an AI system denies a claim, rejects a candidate, prices a customer, flags a worker, or gives medical advice, someone senior must own the outcome.

“The model did it” is not a defence.

It is an admission you lost control.

I made the same point in Who Is Accountable When AI Decides?: AI can recommend, rank, draft, and detect. But it cannot be accountable. That remains a human job.

People need an appeal path

If your AI touches a person’s money, job, health, safety, status, or rights, they need a way to challenge it.

Fast. In plain English. With a human who can fix it.

No appeal path means no trust.

And no trust means the business model eventually breaks.

That is why Trust Is The New AI Battleground. As AI output becomes cheap and everywhere, trust becomes scarce.

Scarce things become valuable.

Do not automate your conscience

Some decisions are too human to hand off.

Hiring. Firing. Policing. Care. Credit. Education. War.

The more serious the consequence, the more human judgment must stay in the loop.

Not as theatre. As authority.

This is the difference between delegation and abdication. In AI Moves Work, Not Magic, I argued that AI does not remove the need for management. It moves the work. It changes where judgment is needed.

Bad leaders use AI to avoid decisions.
Good leaders use AI to make better ones.

Your workers are watching

If AI is just a headcount weapon, expect fear, resistance, leaks, and brand damage.

If AI removes dull work, grows capability, and shares the upside, people may come with you.

“Reskilling” is not a slide. It is a deal.

Who gets trained?
Who gets moved?
Who gets protected?
Who shares in the gains?
Who is told early enough to plan?

That is the actual work.

The better example is not “replace people with bots.” It is redesigning work around new capability. That is why CBA Redesigns Work for AI matters.

The dumb version of AI strategy is cost-out.
The serious version is capability-up.

Truth matters again

AI makes cheap content infinite.

That means trust becomes a moat.

Do not pump out machine-made noise and call it marketing.
Do not fake intimacy.
Do not hide synthetic work.
Do not make customers wonder whether anyone real is home.

Every company will be tempted to use AI to say more, publish more, target more, personalise more, and automate more touchpoints.

Most of it will be junk.
Some of it will be harmful.
A little of it will be useful.

The CEO job is to know the difference.

Institutions still matter

The Pope is not really worried that machines will become human.

He is worried that humans will become machine-like.

Measured. Ranked. Predicted. Optimised. Managed by systems nobody can explain.

That is the deeper warning.

AI can simulate output. It can simulate tone. It can simulate care. It can simulate authority.

But it cannot create trust by itself.

Trust comes from institutions that keep promises, own errors, tell the truth, and protect people when things go wrong.

That is why Synthetic Scale Fails Without Real Institutions is the business version of the same argument.

Scale without trust is fragility
Scale without accountability is risk.
Scale without humanity is just a taller tower.

The Pope says the task is to put “the human person at the center of our choices.”

That is the CEO takeaway.

Share

Use AI.
Move fast.
Compete.

Do not ship systems nobody owns.
Do not hide behind the model.
Do not reduce people to behavioural data.
Do not turn workers into cleanup crews for bad automation.
Do not scale what you would be ashamed to explain.

The Pope’s message is not anti-tech.
It is anti-stupidity.

Build AI that helps people do more, know more, make better calls, and live with more dignity.

Anything else is just Babel with a better user interface.

The AI industry remains obsessed with model performance.

Every major release is evaluated through the same lens. Which model is more intelligent? Which model performs better on reasoning benchmarks? Which model generates better code? Which model achieves the highest score on the latest evaluation suite?

These are useful questions for researchers, developers, and technology teams. They are increasingly less useful for CEOs.

The executive challenge is shifting. The question is no longer simply which model performs best today. The more important question is which AI provider an organisation is prepared to become dependent upon over the next decade.

That distinction may sound semantic. It is not.

Software decisions and dependency decisions are fundamentally different. Software can be replaced. Dependencies become embedded within operating models, governance structures, workflows, and organisational capabilities. Once that occurs, changing providers becomes materially more difficult, regardless of whether a better alternative exists.

Recent developments in the AI market suggest this transition may already be underway.

Enterprise Adoption Is Following A Different Logic

One of the more interesting developments over the past year has been the emergence of different adoption patterns between consumer and enterprise markets.

OpenAI remains the dominant consumer brand. ChatGPT has become synonymous with AI in much the same way that Google became synonymous with search. Consumer awareness, mindshare, and usage remain significant competitive advantages.

Yet enterprise markets have historically followed different rules.

Recent data from Ramp’s AI Index showed Anthropic overtaking OpenAI in business adoption for the first time, with Anthropic reaching 34.4% adoption compared with OpenAI’s 32.3%. Ramp tracks actual spending behaviour across tens of thousands of businesses, making it one of the more useful signals available in the market.

This does not prove Anthropic has built a superior product. Nor does it suggest OpenAI’s position is under immediate threat.

What it does suggest is that enterprise buyers may be optimising for a different set of criteria than consumers.

This would not be unusual. Enterprise technology markets have rarely been won solely through technical superiority. Reliability, governance, integration, vendor maturity, support, security, and long-term viability often become equally important. In many categories, they become more important.

The history of enterprise software is filled with examples where the technically strongest product failed to become the dominant platform.

AI may be entering a similar phase.

Revenue Growth Tells A Different Story

Another useful signal is where revenue growth is emerging.

Anthropic’s reported growth from approximately $1 billion in annualised revenue during early 2025 to more than $30 billion by April 2026 is remarkable. More interesting, however, is the nature of the demand driving that growth.

Much of the value appears to be coming from enterprise use cases. Coding environments, research workflows, legal operations, knowledge management, and internal productivity systems are becoming increasingly important sources of adoption.

These are not casual consumer interactions. They are operational workflows.

That distinction matters because operational workflows create stickiness. Once an organisation redesigns a process around a technology platform, the economics change. The conversation shifts away from feature comparison and toward continuity, governance, and risk management.

Those are the characteristics of infrastructure markets rather than software markets.

The Industry Is Already Behaving Like Infrastructure

The strongest evidence may not come from adoption metrics at all. It may come from how providers themselves are behaving.

OpenAI’s introduction of multi-year Guaranteed Capacity agreements is particularly revealing. The offering allows organisations to reserve long-term AI compute capacity across models and cloud providers, in some cases years in advance.

Capacity reservation is not a typical software construct. It is an infrastructure construct.

Cloud providers sell capacity. Telecommunications providers sell capacity. Utilities sell capacity. Organisations reserve capacity when they expect a service to become operationally critical.

At the same time, AI providers are investing extraordinary amounts into long-term infrastructure commitments. Anthropic’s reported commitment of up to $200 billion with Google is difficult to interpret as anything other than a belief that AI is becoming foundational infrastructure.

The market is gradually moving below the application layer. Models remain important, but increasingly the competitive battle is shifting toward ecosystem control, workflow integration, enterprise governance, and infrastructure scale.

That is typically where long-term winners emerge.

Strategic Implication

The strategic implication is not that organisations should choose Anthropic over OpenAI.

That conclusion would be simplistic and likely wrong.

The more important implication is that executives may be evaluating AI investments through the wrong decision framework.

Most organisations continue to approach AI as a technology procurement exercise. They compare features, benchmark results, and subscription costs. Those considerations matter, but they are not where the largest strategic risks reside.

The bigger question is how AI dependencies are forming across the enterprise.

A Three-Layer Framework For AI Strategy

I increasingly think AI decisions need to be separated into three distinct layers.

The mistake many organisations make is treating all three layers as though they are the same decision.

Subscribe now

Layer One: Models

This is where most executive discussions currently focus. Organisations debate GPT versus Claude, open source versus proprietary models, and performance differences across various tasks.

These decisions are important, but they are also becoming increasingly reversible. Model quality continues to improve across the market, and the gap between leading providers is narrowing.

Most organisations are allocating excessive attention to the layer where switching costs remain lowest.

Layer Two: Platforms

The second layer is significantly more strategic.

Platforms determine how AI interacts with enterprise systems, data assets, workflows, employees, and governance structures. This includes identity management, orchestration layers, agent frameworks, security controls, and integration architecture.

Unlike models, platforms become embedded.

Changing a model may take weeks. Replacing an enterprise AI platform may take years.

This is where dependency risk begins to emerge.

Layer Three: Operating Models

The third layer receives the least attention and may ultimately prove the most important.

AI is not simply changing how work is executed. It is changing how organisations are designed.

Management structures, workforce composition, decision-making processes, customer engagement models, and product development cycles are all beginning to evolve. These are not technology decisions. They are business model decisions.

Boards and executive teams should spend far more time discussing this layer than they currently do.

Three Questions Every CEO Should Ask

The next phase of AI strategy requires a different set of questions.

First, where are we becoming dependent on AI?

Leaders should map the workflows, processes, and decisions that increasingly rely on AI systems. Dependencies create both strategic advantage and strategic risk. Most organisations currently understand the former better than the latter.

Second, which dependencies are reversible?

Not every investment creates lock-in. Understanding which decisions preserve optionality and which create long-term dependence is becoming increasingly important. Switching costs are often invisible until they become operationally painful.

Third, what capabilities must we own?

Every organisation has a small number of capabilities that underpin competitive advantage. The objective is not to own everything. The objective is to ensure that strategically important knowledge, workflows, and decision-making capabilities do not become outsourced by default.

This is particularly important as AI agents become more deeply embedded within enterprise processes. The convenience of external intelligence can easily obscure the strategic value of retaining internal capability.

Share

The Emerging Divide

Over the next five years, organisations are likely to separate into two groups.

One group will continue treating AI as another software category. Their focus will remain on features, pricing, and vendor selection. They will optimise for procurement efficiency and internal productivity.

The other group will recognise that AI is becoming a foundational dependency within the enterprise. Their focus will shift toward strategic control, organisational capability, and long-term optionality. They will treat AI decisions in much the same way previous generations treated cloud strategy, ERP standardisation, or cybersecurity architecture.

However, the most sophisticated organisations will take the thinking one step further.

They will not only ask where they are becoming dependent on AI. They will ask how AI can make them more valuable, more embedded, and more difficult to replace within their own ecosystems.

History suggests the greatest value rarely accrues to those who simply consume infrastructure. It accrues to those who build on top of it.

The organisations that recognise this shift early will not necessarily choose better models.

They are more likely to make better dependency decisions and use AI to become more difficult to displace.

When growth slows, leadership teams tend to reach for the same playbook. They launch new features, cut prices, increase marketing spend, or copy whatever appears to be working for competitors. The logic seems sensible. If a competitor is winning customers, the answer must be to become better than them.

The problem is that every competitor is thinking exactly the same way.

Soon everyone is competing on the same dimensions: price, features, speed, convenience, service. Margins compress. Products become harder to distinguish. Strategic planning turns into a debate about how to win a race someone else designed.

Peter Thiel captured this reality in a single line:

Competition is for losers.

His point was not that competition disappears. His point was that the best businesses stop competing on the industry’s existing terms. They create new ones.

The Trap Of Incremental Improvement

Most strategy work is really optimisation.

Leadership teams spend their time discussing how to improve conversion by 5%, reduce churn by 3%, launch another feature, or move slightly faster than competitors. These are useful questions, but they are not strategic questions.

Strategy starts somewhere else. It asks which assumptions the entire industry is making that no longer need to be true. That is where new market space emerges. Not from being slightly better, but from being fundamentally different.

This is the core idea behind Blue Ocean Strategy, developed by Chan Kim and Renée Mauborgne. Instead of fighting competitors in crowded markets, they argued that organisations should redesign the factors customers value.

The goal is not to win the existing game.

The goal is to create a new one.

The Most Useful Strategy Tool Nobody Uses

Blue Ocean Strategy produced dozens of frameworks. One remains particularly useful because it forces leaders to confront the hardest part of strategy: trade-offs.

The framework is called the ERRC Grid.

Eliminate.
Reduce.
Raise.
Create.

Most leadership teams spend almost all their time discussing the last two. What can we raise? What can we create? How can we add more value? How can we add more capability?

Almost nobody wants to discuss Eliminate.

That is usually where the breakthrough sits.

Every industry accumulates baggage over time. Features nobody uses. Processes nobody questions. Costs everyone assumes are unavoidable. The longer an industry exists, the more these assumptions become invisible.

The ERRC Grid forces leaders to challenge them directly.

The Courage To Be Bad At Something

Southwest Airlines remains the classic example.

The airline eliminated meals, assigned seating, and premium service. Traditional airlines viewed these features as mandatory. Southwest viewed them as costs.

By removing them, Southwest created faster aircraft turnarounds, lower operating costs, simpler operations, and a fundamentally different value proposition. It became intentionally worse at certain things in order to become dramatically better at others.

Most organisations struggle with this idea.

Executives often want the new feature and the old feature. The new service and the old service. The new workflow and the old workflow. Over time, complexity accumulates. Costs rise. Operations become harder. Customers become confused.

The result is a Frankenstein product.

Every meaningful competitive advantage requires sacrifice somewhere else.

That is why every strategy discussion should include one uncomfortable question:

What are we willing to be terrible at?

If you are trying to be excellent at everything, you are probably not creating a Blue Ocean. You are simply adding complexity.

Use this prompt to challenge the assumptions in your industry.

Prompt: Blue Ocean ERRC Grid

The goal is not to improve the existing market.

The goal is to redesign it.

Share

The Executive Test

Before approving any new feature, service, initiative, or investment, ask one question:

What are we eliminating?

If the answer is nothing, there is a good chance complexity is growing faster than value.

The strongest strategies are rarely additive. They are selective. They force leaders to decide what not to do, what not to build, and which customers not to serve. That discipline is often what creates the advantage.

The most successful companies often look strange when they first emerge because they are not trying to win the existing game.

They are building a different one.

Below are two governance tools that force the hardest part of Blue Ocean Strategy: making deliberate trade-offs. Most leadership teams understand the framework. Far fewer have the discipline to apply it.

Subscribe now

The Trap: Gold Plating

Most executives are afraid to eliminate anything.

Every strategic initiative arrives as an addition. New products, new channels, new reports, new features, new processes, new approvals. Very little ever leaves.

Over time, complexity compounds. Costs rise. Decision-making slows. Customer experiences become cluttered.

Blue Ocean Strategy only works when leaders are willing to remove things, not just add them.

The following prompts are designed to force that discipline.

Follow-Up Prompt 1: The Anti-Feature List

Use this to identify and remove complexity.

The objective is to separate genuine customer value from inherited complexity.

Follow-Up Prompt 2: The Non-Customer Tier

Use this to identify growth that competitors cannot see.

Most growth strategies focus on stealing customers from competitors.

Blue Ocean strategies focus on people who are not customers yet.

That is often where the biggest opportunities sit.

The Executive Takeaway

Competition feels productive because it is measurable.

You can track market share. You can benchmark features. You can compare pricing. You can watch competitors. Every board meeting can be filled with competitor analysis and market updates.

Creating a new market space is much harder because there is no scoreboard.

Which is exactly why it works.

Most organisations spend their energy fighting over existing demand. The best organisations redesign the market so they no longer have to.

That is the real lesson behind Blue Ocean Strategy.

Stop asking how to beat your competitors.

Start asking how to make them irrelevant.

The Workforce Is Changing Faster Than The Workplace

Some people now instinctively open ChatGPT before Google.

Others still use AI occasionally, mostly for experimentation, then return to their normal workflows.

That gap matters more than it appears.

One employee starts every draft with AI. Another still begins with a blank page. One asks AI to summarise reports, pressure-test ideas, structure meetings, and accelerate research. Another still treats AI as an occasional assistant layered onto an unchanged workflow.

At first glance, the difference feels small. Over time, it may become enormous.

Because the real AI divide may not ultimately be technical. It may simply separate people who rebuild their habits around AI from those who continue operating largely as they always have.

The data increasingly points in this direction.

Stanford’s 2025 AI Index found that 78% of organisations now report using AI in some form. Yet McKinsey found only 1% of executives describe their companies as “mature” in AI adoption, meaning AI is fully integrated into workflows and generating meaningful business outcomes.

That gap matters.

It suggests most organisations now have access to AI tools, but relatively few have deeply changed how work actually happens.

Gallup’s workplace research tells a similar story. Around 40% of employees report using AI at work at least occasionally, but only a small percentage use it daily. In remote-capable knowledge work, habitual use rises sharply, particularly among leaders and highly digital teams.

The implication is becoming difficult to ignore.

Access to AI is spreading rapidly. Behavioural integration is not.

Technology Changes Faster Than Habits

Organisations consistently underestimate how difficult habits are to change.

Corporate history is full of technologies that promised behavioural transformation but were mostly absorbed into existing routines.

Email did not eliminate meetings. In many organisations, it increased them.

Slack did not eliminate email. It simply created another communication layer.

Dashboards did not eliminate PowerPoint. Most companies still converted dashboards into presentations for executive meetings.

Generative AI may follow the same pattern.

Many organisations currently describe themselves as “adopting AI” when what they are really doing is layering AI tools onto fundamentally unchanged workflows, management structures, approval systems, and communication habits.

The software changes immediately. Behaviour changes slowly.

The interesting part is not that AI tools are spreading quickly.

It is that repeated AI usage appears to be changing behaviour itself.

That shift may become one of the most important competitive dynamics of the next decade, particularly for organisations still treating AI as a software rollout rather than a behavioural transition.

The rest of this article explores where the real divide may emerge, why some companies are already pulling ahead, and what leaders may still be underestimating.

Read more

The Productivity Boom Nobody Is Questioning

A junior finance analyst can now generate a polished board-level variance summary in under five minutes using AI. Five years ago, that same analyst may have spent half a day tracing transactions, checking assumptions, investigating anomalies, and understanding what actually drove the numbers. The new process is dramatically faster. It may also produce weaker financial judgment over time.

That tension sits underneath much of the current excitement around AI inside finance functions. Most executive conversations understandably focus on productivity gains. Faster close cycles, leaner reporting teams, automated reconciliations, AI-generated commentary, and improved forecasting all present compelling operational and economic advantages. If AI can safely compress hours of analytical work into minutes, every finance organisation will eventually face pressure to deploy it.

The more difficult question is what happens once large portions of the underlying work disappear.

Finance Functions Run On Judgment

Finance functions do not ultimately operate on software. They operate on judgment. Experienced finance leaders develop an instinct for anomalies, inconsistencies, fragile assumptions, and operational risks that rarely appears explicitly inside systems or dashboards. Much of that judgment forms through repetition and exposure.

Analysts learn by tracing discrepancies across systems. Accountants learn by manually resolving broken reconciliations. Teams build commercial intuition by repeatedly working through situations where the numbers do not initially make sense.

Historically, much of this work looked inefficient. In reality, it was also training.

That distinction matters because AI is arriving at the exact layer where many organisations unknowingly developed future financial leaders.

Finance Has Seen Versions Of This Before

Most finance teams have already experienced smaller versions of this dynamic through spreadsheet dependency.

A model becomes operationally critical over time until only one or two people fully understand the underlying logic. Everyone else learns the workflow but not necessarily the assumptions, calculations, or structural weaknesses embedded underneath it. The spreadsheet continues producing outputs while organisational understanding gradually narrows.

The problem only becomes visible when something changes. A key employee leaves. A formula breaks. A reporting anomaly appears. A business assumption shifts. At that point, the organisation discovers it can still operate the process, but fewer people can confidently explain how the model actually works.

That is organisational memory loss in practice. Not dramatic collapse, but a gradual separation between producing outputs and understanding how those outputs are produced.

The process survives. The understanding becomes concentrated.

AI Changes The Development Path

Generative AI may scale this dynamic significantly because it reaches much further upstream than previous forms of enterprise automation.

Earlier generations of finance technology primarily automated workflows, processing, and administration. AI increasingly automates interpretation, synthesis, explanation, and analysis. Those activities have historically been where junior finance professionals developed judgment.

This creates a potentially uncomfortable paradox.

The first generation of AI-enabled finance teams may become highly productive before they become deeply experienced.

At first, the indicators look entirely positive. Reports improve. Commentary becomes more polished. Teams move faster. Junior staff contribute earlier. Executives receive cleaner summaries and faster responses.

Operationally, the transformation appears successful because, in many respects, it is successful.

The challenge is that highly polished outputs can obscure weakening capability underneath.

The first generation of AI-enabled finance teams may become highly productive before they become deeply experienced.

Subscribe now

The first wave of AI adoption has been dominated by productivity gains. The second wave will be defined by organisational consequences that most leadership teams still underestimate.

That is the focus of For Every Scale.

I write for CEOs, CFOs, COOs, boards, and executive teams navigating how generative AI is changing operational capability, management structures, and decision-making inside large organisations.

If this article resonated with you, subscribe for analysis on the strategic and organisational implications of AI beyond the hype cycle.

The Friction Was Doing More Than People Realised

A surprising amount of financial judgment historically came from doing slow, difficult work manually before automating it.

Analysts built instinct by repeatedly encountering inconsistencies, exceptions, and unexplained variances. Over time, they learned not just how to produce numbers, but how to interrogate them.

Generative AI may remove portions of that developmental layer. Not intentionally, but structurally.

There are already signs of this pattern emerging inside highly automated accounting environments. Researchers studying automated accounting workflows documented situations where staff struggled to manually reconstruct processes once automation systems were removed.

The systems operated effectively while the automation layer functioned normally. The vulnerability only became visible once people needed to work without it.

The problem with removing all friction from knowledge work is that friction is often where understanding comes from.

The Leadership Question Is Changing

For CFOs and COOs, this changes the leadership question entirely.

The challenge is no longer simply determining how much work AI can automate. The harder question is how the organisation continues developing financial judgment once many of the hard parts disappear.

That is not purely a technology problem. It is an organisational design problem.

Most executive dashboards are designed to measure productivity, throughput, cost reduction, reporting speed, and close-cycle compression. Those metrics matter and should matter. However, they do not necessarily indicate whether the organisation is still producing future experts.

They do not measure whether junior finance professionals are developing the judgment required to operate effectively under ambiguity, stress, or abnormal conditions.

Boards do not simply want accurate reporting. They want confidence the organisation understands why the numbers are accurate.

A Framework For Leadership Teams

The practical test for leadership teams is not whether AI saves time. It is whether the organisation has a plan to replace the learning that the saved time used to create.

A useful starting framework is what I think of as the Capability Preservation Test.

1. What work are we automating?

Not the process name. The actual judgment being removed from human practice.

A reconciliation process may also be teaching anomaly detection. A forecasting workflow may also be building commercial intuition. A reporting task may also be developing pattern recognition.

2. What capability did that work used to build?

Many repetitive finance activities were unintentionally serving as apprenticeship systems.

Before removing the work entirely, leadership teams should identify what forms of judgment were historically being developed through repetition and exposure.

3. Where will that capability now be developed?

This is where many organisations currently have a blind spot.

If AI removes large portions of manual analytical work, where does the next generation of financial instinct come from? What replaces the developmental pathway that previously existed?

4. How will we know people still have the capability?

Most organisations test process compliance. Far fewer test operational judgment.

That may need to change.

Manual reconstruction exercises, anomaly reviews, edge-case simulations, and model challenge sessions may become increasingly important in highly automated environments.

5. Who owns the capability, not just the system?

System ownership is not the same thing as capability ownership.

Someone inside the organisation needs to be accountable for ensuring critical financial judgment continues developing over time, even as automation increases.

The Risk That Arrives Slowly

The real long-term risk may not be hallucinations or bad outputs. It may be a gradual erosion in the organisation’s ability to independently understand its own operations.

Not because the systems stop working, but because fewer people are repeatedly exposed to the friction through which deep financial judgment historically formed.

The irony is that organisations moving fastest with AI may not notice this dynamic until years later because the early indicators look exactly like success. Productivity improves. Reporting accelerates. Teams become leaner. Outputs become more sophisticated.

Until the organisation encounters something the system was never designed for and discovers too few people still know how to think through the problem from first principles.

This is first-order problem solving: treating the visible failure instead of the system that created it.

Taiichi Ohno’s rule was simple:

The mistake most organisations make is assuming the root cause sits close to the incident itself. In practice, the opposite is usually true.

What looks like a technical failure at Level 1 is often a process failure at Level 3 and a leadership failure at Level 5.

A missed deadline is rarely caused by “slow engineering.” A customer escalation is rarely caused by “poor communication.” An outage is rarely caused by “human error.”

Those are surface-level manifestations of deeper operating design problems.

The five-level pattern

A typical failure chain looks like this:

Problem: We missed the Q1 shipping deadline.

1. Why did we miss the deadline?
   Engineering delivery slipped.

2. Why did engineering delivery slip?
   Requirements kept changing.

3. Why did requirements keep changing?
   Sales kept committing to new features during the quarter.

4. Why was Sales doing that?
   Their incentives rewarded revenue generation, not delivery feasibility.

5. Why were incentives structured that way?
   Leadership optimised compensation for growth without integrating operational constraints.

That fifth layer matters because it changes the corrective action completely.

A Level 1 fix sounds like this:

“Engineering needs to execute faster.”

A Level 5 fix sounds like this:

“Redesign commercial incentives so feature commitments require delivery sign-off.”

One treats symptoms. The other changes the system.

The “human error” trap
Many organisations stop their root cause analysis at the first socially convenient answer:

“Someone forgot.”
“Someone missed the check.”
“Someone made a mistake.”

That is not root cause analysis. That is blame with administrative language.

If your investigation ends with “human error,” the process design is probably incomplete.

Humans forget things. Humans skip steps. Humans get distracted. Good operating systems assume this in advance.

Toyota’s manufacturing philosophy was built around this principle. Instead of trying to create flawless humans, they designed systems that prevented predictable mistakes from occurring in the first place.

That distinction matters because organisations that punish individuals for process failures usually create two downstream problems:

1. Employees hide issues earlier.

2. The underlying failure mechanism survives untouched.

The result is recurring operational instability.

The executive rule

A useful leadership rule is this:

Never fire a person for a process failure you designed.

That does not eliminate accountability. It relocates accountability upward, toward system ownership.

If the process allows a predictable mistake to occur repeatedly, leadership owns the process.

This is why mature operational cultures focus less on blame and more on controls, incentives, interfaces and decision architecture.

The goal is not to find who failed.

The goal is to understand why the system permitted the failure.

Use this prompt to run a proper root cause analysis.

Prompt: Toyota 5 Whys Analysis

The next problem: recurrence

Even when organisations identify the root cause correctly, many still fail because they do not redesign the process afterward.

They conduct the post-mortem, document the findings, circulate the slides, then leave the workflow unchanged.

That guarantees recurrence.

Toyota addressed this using “Poka-Yoke”: mistake-proofing mechanisms that either prevent the error entirely or make it immediately visible.

A strong operational process should not rely on memory, heroics or vigilance.

It should make the correct behaviour automatic.

Examples:

A deployment pipeline that blocks production release without test coverage
A CRM that cannot progress without mandatory data fields
A procurement workflow that requires dual approval above spending thresholds
A board paper template that forces the recommendation onto page one

The pattern is always the same:

Do not ask people to remember critical controls.
Engineer the controls into the system.

Use this prompt to redesign the process itself.

Subscribe now

Follow-Up Prompt: Poka-Yoke Process Design

The culture problem

There is another reason most post-mortems fail:

People become defensive before the meeting even starts.

The moment employees believe the process is about blame allocation, the organisation loses access to accurate information. Teams minimise ownership, hide uncertainty and protect themselves politically.

That destroys operational learning.

High-performing operational cultures do the opposite. They separate accountability from humiliation.

The rule is simple:

That framing matters because operational transparency depends on psychological safety. Teams only surface weak signals early if they believe the organisation wants truth more than scapegoats.

The leadership move is important here.

The senior leader should absorb ownership of the Level 5 issue personally. That lowers organisational temperature immediately because it signals that the objective is diagnosis, not punishment.

Use this prompt to open a blameless post-mortem properly.

Follow-Up Prompt: Blameless Post-Mortem

The executive takeaway

Most operational failures are not random.

They are system outputs.

If the same issue keeps recurring, the organisation is probably treating symptoms instead of redesigning the mechanism producing them.

Rebooting the server may restore service.
It does not explain why the outage happened.

Replacing the employee may remove the individual.
It does not remove the structural condition that created the mistake.

The leadership discipline is to keep digging until the organisation reaches the design flaw underneath the visible incident.

That is the purpose of the 5 Whys.

Not to explain failure.

To expose the system that generated it.

After I published my recent piece on AI and “the Clankers”, a previous work mate commented: “It is a fine article, but does bear the hallmarks of being written substantially with AI.”

It was a fair observation.

AI absolutely helped smooth parts of the piece. But the comment stayed with me for a different reason.

Because it exposed something much larger now happening inside organisations.

Subscribe now

AI is weakening many of the traditional proxies we used to identify talent.

For years, companies rewarded people who could produce polished outputs. Strong presentations. Sharp written communication. Clean synthesis. Structured strategy language. Entire careers were built around the ability to look competent in information-rich environments.

The problem is that AI is becoming very good at all of those things.

Which means leaders now face a harder question:

How do you identify actual capability once polished output becomes cheap?

This is already happening. Executives are reading immaculate briefing papers and wondering whether the person behind them can actually think. Managers are reviewing beautifully structured strategy decks and quietly questioning how much original judgment sits underneath the formatting and fluency.

The old signals are weakening.

That matters because most organisations were designed around observable output. Promotions, influence, and credibility often flowed toward the people who sounded the smartest in the room.

AI changes the economics of that competence.

A mediocre thinker with AI can now produce work that superficially resembles the output of someone much stronger. Not equivalent. But close enough to create management noise.

The premium shifts away from production and toward discernment.

Can this person make good decisions under uncertainty?
Can they identify what matters before the data is obvious?
Can they ask better questions than everyone else?
Can they explain why the machine is wrong?
Can they see second-order consequences?
Can they earn trust?

These were always valuable traits. AI simply makes them easier to distinguish from performative competence.

Ironically, this may improve some organisations over time.

For years, many companies accidentally rewarded presentation fluency over insight, activity over judgment, polish over originality, and information control over decision quality. AI compresses the value of those things.

That does not mean expertise disappears.

It means expertise becomes harder to fake.

A useful analogy came back to me after that comment.

If someone eats an extraordinary meal, they might say:

“You must have a great stove.”

But experienced chefs know the stove is not the meal.

The stove matters. Great tools absolutely matter. But tools amplify capability unevenly. A great chef with an average stove still outperforms most amateurs with a commercial kitchen.

AI works the same way.

A weak strategist with ChatGPT is still a weak strategist. They simply become faster at producing plausible-looking work. Meanwhile, exceptional operators become dramatically more effective because the machine removes low-value friction around execution.

Share

That distinction is going to matter enormously for leadership teams over the next few years.

The answer is not to ban AI. That will fail, and in most roles it would be a strange objective anyway.

But leaders do need moments where the tool is removed.

My kids still do some assignments and exams by hand. Not because handwriting is the future, but because sometimes you need to see what is actually in the student’s head.

Workplaces need the equivalent.

Not permanently. Not theatrically. But deliberately.

Ask someone to explain their recommendation without the deck. Give them a live problem and watch how they reason. Ask what they changed after using AI. Ask which part of the machine’s answer they rejected and why.

The test is no longer whether someone can produce polished output.

The test is whether they understand it well enough to defend, adapt, challenge, and improve it.

That creates a practical leadership shift.

Use AI for production.

Test humans for judgment.

The board paper can be AI-assisted. The discussion should not be. The strategy deck can be polished with tools. The executive presenting it should still be able to handle ten minutes of unscripted challenge.

This is how organisations cross the chasm.

They stop pretending AI use is the issue.

The real issue is whether the human has become more capable, or merely more fluent.

Leave a comment