Tuesday, midmorning, in the capital of a suspiciously hexagonal country I dare not name. A board meeting I cannot identify either. They never let me.
The board is halfway through the AI update. Slide 14 offers a three-tier maturity model in reassuring navy blue. Slide 15 is a heat map with sixteen boxes, twelve of them amber, the official color of corporate indecision. Slide 16 announces “cultural readiness” in a proprietary font that clearly came with billable hours.
The CFO has taken his pen (a Montblanc, for sure) apart and put it back together so many times that he now understands its operating model better than the company’s AI one. A nonexecutive director with great corporate hair and a two-tone Rolex Datejust (for sure), asks, very gently, whether the ROI figure on slide 22 is the same one presented last November. There is a small pause. The chief of staff explains that the methodology has since been refined.
The pen comes apart again.
I have sat through variations of this meeting from Brussels to Paris, Ghent, Austin, Amsterdam. A virtual one in Casablanca. The choreography barely changes, neither does the language. It usually begins with little “oopses.” The word is doing heroic work. It turns deliberate executive avoidance into something that appears to have happened by accident, like spilling coffee on the minutes (if there are any minutes, they are a dying breed).
“Oops, we clearly need to invest. Let’s make sure somebody else moves first.”
“Oops, our people need acculturation. We should probably run a survey.”
“Oops, where is the ROI? Could someone benchmark us against a peer set?”
“Oops, before we approve a budget, could we give everyone free ChatGPT, Claude, Perplexity, and Grok accounts and see what happens?”
It never stops at four. “Oops, our data isn’t ready,” now entering its third consecutive year in most organizations. “Oops, we need a Chief AI Officer,” reporting to whom exactly, and paid from whose budget? “Oops, the EU AI Act is watching,” as though it slipped into Brussels overnight wearing a balaclava. “Oops, our competitors are ahead,” when they are usually just as lost, but considerably louder about it on LinkedIn.
Then come the very expensive oopses. “Oops, perhaps we should buy a startup,” which generally means overpaying for it in Q4 and suffocating it by Q2. “Oops, legal has questions,” because legal has always had questions and this remains, inconveniently, legal’s job. And the eternal favorite: “Oops, perhaps we should wait for the next model release. It might change everything.” That sentence has kept several transformation programs safely stationary through three generations of models.
The repertoire can fill ninety minutes without ever approaching a destination. Then the deck ends, somebody requests a revised roadmap, and the meeting breaks for lunch.
Nobody asks the one question capable of ruining the mood: where do you want to get to?
You remember the scene. Alice (the one from Wonderland, not the song) reaches a fork in the road and looks up at the Cheshire Cat, grinning in a tree, half tail, half smile, with the serene confidence of someone who will never be asked to produce an implementation plan.
“Would you tell me, please, which way I ought to go from here?”
The Cat, arguably the most honest consultant in English literature, answers without a deck:
“That depends a good deal on where you want to get to.”
Alice replies that she does not much care where.
“Then it doesn’t matter which way you go,” says the Cat.
“So long as I get somewhere,” Alice adds.
“Oh, you’re sure to do that,” says the Cat, “if you only walk long enough.”
Corporate Alice has been walking for years. The pilots multiply. The steering committee still meets. Copilots have arrived in every inbox, and enough amber heat maps have been produced to wallpaper the executive floor. We have certainly arrived somewhere.
The only mystery is where.
Alice is us. The fork is 2026. The Cat has seen the acculturation slide and is (again) losing patience.
The AI margin and the theater performed there
I call the whole performance the AI margin. It is the narrow strip where an organization can look heroically busy while keeping every consequential choice safely out of reach.
Pilots breed there. They produce encouraging results for the six and a half people who were already convinced, then expire somewhere between procurement and the operating model. POCs settle comfortably inside IT, where everybody can admire them without changing how the company works. Training programs teach people to prompt more fluently while their roles, targets, and decision rights remain untouched. Once a month, the steering committee eagerly reopens last month’s deck, moves three amber boxes a few pixels to the right, and asks for more granularity.
The margin is comfortable. It presents beautifully to a board, and every outcome arrives wrapped in enough caveats to be defended later. It can quietly swallow two or three million euros a year in a mid-cap through licenses, consultants, workshops, sandboxes, hackathons, governance meetings, and the ceremonial catering that accompanies them (ah the joy of lukewarm bubbles and suspicious canapés) . The operating model survives mostly unharmed. Plausible deniability comes bundled at no extra charge.
Brian Solis, head of global innovation at ServiceNow, has been attacking this reflex with admirable energetic impatience. Most leadership teams point AI at the company they already have and ask it to reproduce the same output faster, more cheaply, and with fewer people. The savings fit neatly into a quarterly dashboard, so the exercise acquires the reassuring smell of stale strategy.
The result is -at best- yesterday’s company with better response times. A board can automate that company and keep the corner office right up to the morning somebody else arrives with tomorrow’s.
Stanford economist Erik Brynjolfsson has spent years putting evidence beneath that discomfort. In The Turing Trap, he describes the economic temptation built into automation. Replacing human work produces visible savings, immediate metrics, and a result the CFO can place in a spreadsheet. Augmentation asks harder questions about capabilities, products, roles, and the distribution of value. It requires imagination before arithmetic.
In a July 2026 conversation with MIT Sloan Management Review, Brynjolfsson returned to the distinction with fresh labor-market evidence. Employment held up better in occupations where AI extended human capability. Roles exposed primarily to substitution were already showing pressure. One set of decisions increased the economic value of human judgment. The other gradually removed it from the equation.
Later that month, Brynjolfsson helped organize We Must Act Now, a statement signed by more than 200 economists and AI researchers, including 16 Nobel laureates and leading economists from OpenAI and Anthropic. They called for urgent work on the incentives, guardrails, and institutions needed to steer AI toward complementing people and spreading the gains.
Once more than 200 economists and 16 Nobel laureates climb onto the same roof and pull the fire alarm, somebody downstairs should probably look up. The alarm reached the boardroom. The minutes record a thoughtful discussion. Then the chair moved to item seven (ergonomic office chairs).
The board is not off the hook
The CEO has enjoyed enough attention. The CEO is easy. One person, one office, one photograph in the annual report looking thoughtfully toward a horizon just outside the frame.
The executive committee deserves its turn, because that is where AI conversations go to be domesticated (and/or die). The CFO wants a number by Q3. The COO wants a quiet pilot inside one function, preferably with a clearly marked emergency exit. The CHRO wants a policy document, an acculturation program, and something reassuring about skills. The CIO wants an architecture capable of surviving the vendors’ next round of mergers. The CMO wants a sparkling story for the annual report.
Every request is perfectly reasonable. After ninety minutes, all the dangerous and adventurous edges have been sanded away. Finance, operations, HR, IT, and communications can live with the result.
So can every competitor.
The board sits one step farther away and, in many European mid-caps I encounter, one degree less curious. AI appears once a year in the risk section, usually parked next to cybersecurity and resilience. It gets a paragraph. Occasionally, during a particularly adventurous year, it gets a page. A director asks whether the company already has ChatGPT, and what it costs. The chair thanks management for the comprehensive update. The agenda moves on.
That comfortable distance has become a governance risk of its own. Citi Institute now states that agentic AI creates operational, reputational, and regulatory exposures for which boards carry fiduciary responsibility. Its message to directors is unusually blunt: “Boards can no longer treat AI as a technology initiative delegated only to the chief technology officer.”
Fasken connects directors’ own use of AI directly to their duties of care and confidentiality. KPMG and INSEAD’s April 2026 governance principles place strategy, technology, security, workforce transformation, human accountability, trustworthy AI, and the board’s own working practices firmly on the boardroom table (where it should have been in the first place).
Nasdaq’s 2026 governance guidance dives right into the daily mechanics. Oversight now includes the way directors use AI to summarize board materials, analyze sensitive information, prepare questions, and support their own decisions. The board pack is now in scope. So is the director who drops page 127 into an unsecured consumer chatbot five minutes before the meeting and asks it to produce three intelligent questions.
A board can live a long and perfectly happy life without understanding transformer architecture. They still need enough fluency to look management in the eye and ask:
“Where is AI running in this company? Which decisions does it influence? What data crosses its boundaries? Which vendors sit inside the chain? Who owns the outcome when it fails? And then the question with teeth: do we have the guts and the building blocks to let AI change how this company actually works? Can intelligence become part of our operating system, shaping decisions, workflows, products, and customer promises every day? What are we prepared to redesign, unlearn, or kill to make that happen?”
The operating system here is the messy machinery of the company itself: how it decides, allocates, builds, sells, learns, rewards, and stops. AIQ begins when intelligence reaches that layer. Management may have a clean inventory of systems, approved vendors, and compliant use cases while having no answer to the question with teeth. That leaves the board governing a catalog of tools as the operating model carries on undisturbed.
If another roadmap is the only answer, the board has agreed to be surprised. Surprise is delightful at birthdays. In governance, it tends to arrive with lawyers.
The Cheshire question hiding in plain sight
Back to Alice at the fork.
The Cat’s answer exposes the rusty anatomy of most corporate AI conversations. When direction goes missing, motion takes over. The road starts impersonating the strategy. The pilot becomes the objective. The vendor deck supplies the vision. License activation passes for cultural change. The board approves the whole thing because motion renders beautifully in a quarterly update.
Before the investment slide, before the acculturation program, before the ROI calculation built on assumptions wearing decimal points, one question belongs at the center of the table:
“What should now be possible in this company that was impossible, unaffordable, or simply invisible before?”
That is the AIQ question.
AIQ stands for artificial intelligence quotient. At SXSW 2026, Brian Solis placed it alongside IQ, emotional intelligence, social agility, and the much rarer quality of genuine self-awareness. I have kept returning to that diagram because it exposes a confusion spreading through companies at remarkable lethal speed.
An executive can prompt beautifully and still have no idea what the company should become.
Prompting is basic literacy. Producing a plausible memo in thirty seconds is already losing its magic. Automating a support ticket is plumbing. Buying licenses is procurement. AIQ begins when a leadership team can look at its business and name, in fresh sentences, the capabilities that have suddenly moved within reach.
Could a customer promise that was once ruinously expensive become affordable? Could a quarterly decision become continuous? Could a product adapt to one person at a scale of millions? Which risk could become visible before it turns into an incident? Which market could a team of eight suddenly enter? Which month-long process could collapse into an afternoon while preserving the judgment that made it valuable?
Those questions pull AI out of the tool layer and into the company’s operating system. They touch workflows, decision rights, products, incentives, skills, margins, and the promises made to customers. They also force the executive committee to reveal how far its imagination reaches beyond the software catalog.
An executive committee that cannot answer them without opening a vendor deck has a purchasing program wearing a lanyard marked STRATEGY (about as valuable as snake oil in a LVMH tube). Close the deck. Hide the logos. Put the maturity model away. Ask the question again. The silence that follows may be the most accurate maturity assessment you see all year.
Faster and cheaper gets you to the starting line
A lazy consensus is forming around AI. It will do a plethora of things, produce emails, reports, code, presentations, and redundancies at greater speed. If that is the full extent of your ambition, you can stop reading and go approve the Copilot invoice.
Give it three years and every serious competitor will have access to roughly the same models, running on roughly the same infrastructure, making roughly the same promises about productivity. The starting line will be crowded with companies congratulating themselves for roughly having reached it.
The useful question is what your company can now do that previously sat beyond the edge of the map.
Moderna moved early and wide. Its work with OpenAI began in 2023, and ChatGPT Enterprise subsequently spread across thousands of employees and functions. Within two months of one rollout, employees had created 750 custom GPTs. Forty percent of weekly active users had built one themselves, and the average user was clocking 120 conversations a week. A month of my personal usage looks restrained by comparison, and I write for a living.
The 750 GPTs demonstrate enthusiasm. Dose ID shows where things become more interesting. The tool analyzes clinical data, searches and cites sources, produces visualizations, and helps clinical teams reason through dose selection. Human experts still make the decision, but the range and speed of judgment available to them have changed. The work has begun moving underneath the organization chart, which is now doing its best to look as though this was the plan all along.
Novo Nordisk went bigger, more methodically, and considerably more Danish about it. In April 2026, the company announced a strategic partnership with OpenAI spanning drug discovery, clinical development, manufacturing, and commercial operations. Workforce upskilling, governance, data protection, and human oversight were built into the program from the outset. There was no ceremonial chatbot released into the lobby to see whom it might impress.
Inside its research organization, a governed AI system built on Microsoft Azure has moved some analyses from weeks to minutes. Teams that previously had the capacity to investigate five to ten promising ideas per quarter can now evaluate more than fifty. In May 2026, Novo told Reuters that it was targeting a reduction of up to two-thirds in the time required to bring drugs to market.
The spreadsheet will duly and happily record the savings. Patients may experience (and survive) the part that matters: a molecule reaching them months/years/decades earlier.
Duolingo chose a louder route and discovered what happens when an AI metric meets human incentives. In April 2025, CEO Luis von Ahn declared the company “AI-first.” Contractors doing work that AI could handle would gradually be phased out. AI usage would influence hiring, resource allocation, and performance reviews. The announcement produced exactly the sort of public reception one might expect when a cheerful green owl appears to be carrying a scythe.
Employees pushed back, users protested. More importantly, the metric began encouraging the wrong behavior. By April 2026, von Ahn acknowledged that the company had dropped AI usage from performance evaluations after employees asked, “Do you just want us to use AI for AI’s sake?” The answer, eventually, was no.
That reversal deserves more than a cheap victory lap. Duolingo set an incentive, watched it distort behavior, and removed it in public. The company kept AI at the center of its strategy while abandoning compulsory enthusiasm as a management technique. Plenty of executive teams would have commissioned a culture survey and quietly renamed the metric.
Taken together, these companies offer three answers to the same boardroom question: what, exactly, is AI for here?
In April 2026, Harvard Business Review described the choice as one between extracting bottom-line gains through automation and creating top-line growth by augmenting people. Brian Solis draws a similar distinction between iterative AI and innovative AI. One improves the company already sitting in front of you. The other asks what company could be sitting there five years from now.
Efficiency enters the meeting carrying a flashy color-coded spreadsheet. Reinvention arrives with hypotheses, awkward questions, and a shaky time horizon that makes the quarterly calendar nervous. Guess which one gets the CFO’s chair.
Every company can buy access to the models. The scarce resource sits in the corner deciding which work should disappear, whose judgment should grow, and what the company intends to become once faster and cheaper are no longer interesting.
That scarce resource is AIQ.
The four risks nobody puts on the risk slide
The enterprise AI risk slide is usually immaculate. Hallucinations, copyright, bias, privacy, cybersecurity. Five tasteful icons, three amber status bubbles, and a footnote in eight-point type confirming that legal has been consulted.
All legitimate. Also heartbreakingly incomplete. The risks developing inside the company’s habits, language, and collective brain are harder to insure, harder to assign to legal, and awkward to place beneath a green checkmark. Consequently, they tend to disappear somewhere between the steering committee and slide 17.
Commoditization. Most companies are connecting the same handful of models to the same cloud platforms, feeding them the same public information, and deploying them through the same consulting accelerators. The result can be fast, competent, professionally punctuated beige.
In July 2026, Pangram published an analysis of 1,002,627 social media posts collected through its browser extension. More than 40 percent of the long-form LinkedIn posts it scanned were flagged as fully AI-generated. Treat that figure as an AI detection company’s estimate, rather than tablets brought down from Mount Sinai. The dataset came from users who opted in, and the classification came from Pangram’s own detector. Even with those caveats, the direction is difficult to miss. LinkedIn is beginning to read as though one enormous intern was asked to make everybody sound more thought-leadery.
Now bring that sameness behind the firewall. The customer presentation, the strategic recommendation, the product roadmap, and the CEO’s annual letter all pass through the same intelligence layer. Distinctive language goes first. Distinctive thinking tends to follow it out of the building, annoyingly fast.
Give AI an original point of view and it can sharpen it. Give it a vacuum and it will return beautifully formatted vacuum.
Cognitive addiction. Companies are currently celebrating active seats, prompt volumes, adoption curves, and daily usage. Employees are encouraged to consult AI constantly, preferably in ways that can be measured and placed on slide 14. Very few dashboards ask whether people can still perform the task when the tool is unavailable, wrong, or merely very convincing.
A 2026 study of 504 working adults in Shanghai found significant associations between AI overreliance and dependency, anxiety, fear of missing out, addictive behavior, cognitive overload, and technostress. Surprised? The design was cross-sectional and based on self-reported data, which leaves causality outside its reach. Still, the cluster deserves attention. Anxiety and technostress were among the strongest predictors of overreliance.
There is a rather elegant corporate loop hiding in there. Management creates pressure to use AI. Employees use AI to cope with the pressure created by the AI program. Usage increases, the adoption dashboard turns green, and somewhere a transformation officer receives a bonus. You reach for the tool. The tool becomes the reflex. Six months later, writing the difficult paragraph yourself feels strangely inefficient. Then the same thing happens to analysis, judgment, and disagreement. Dependency rarely announces itself with a system alert. It arrives as convenience and puts its feet under the desk.
Slop. Slop does not remain on LinkedIn. It has an employee badge and has already entered the building. Reports are written by AI, reviewed by AI, summarized by AI, and forwarded to executives who use AI to summarize the summaries. The original evidence is resting somewhere in SharePoint beneath FINAL_v7_reallyFINAL.docx. Nobody has read it. Nobody can remember who requested it. Every layer of the sandwich is bread.
The prose remains smooth. The headings are correctly nested. The recommendations sound eminently reasonable because reasonable is the average temperature of everything the model has ever absorbed. One executive copies a sentence into the board pack. Another asks Copilot to prepare questions about it. A third asks for an executive summary of the resulting discussion. By Friday, the company has generated eighteen pages of material without adding a thought.
This creates a very specific governance problem. Decisions retain their audit trail while losing their intellectual author. When a regulator, customer, employee, or shareholder eventually asks why a decision was made, because Copilot thought it sounded reasonable will struggle to carry the room.
Atrophy. This is the word that sells expensive conference tickets, so it needs to be handled with silk gloves.
A Microsoft Research and Carnegie Mellon study presented at CHI 2025 surveyed 319 knowledge workers about 936 real tasks completed with generative AI. Greater confidence in the AI was associated with less critical thinking. Greater confidence in one’s own abilities was associated with more. The tool also changed where critical effort was spent, shifting it toward checking information, integrating answers, and supervising the task.
Then came the MIT Media Lab essay-writing study. Fifty-four participants completed the first three sessions using an LLM, a search engine, or no external tool while researchers monitored brain connectivity with EEG. The LLM group showed the weakest connectivity, alongside lower recall and a weaker sense of ownership over what they had written.
The viral number from that study was a reduction of up to 55 percent. What the paper actually reported was up to 55 percent lower total connectivity magnitude in specific low-frequency networks compared with the brain-only group. Somewhere between the preprint and LinkedIn, this became 55 percent less brain activity, as though more than half the brain had quietly gone out for lunch. That translation is nonsense (and spectacular framing: well done).
The study remains small, task-specific, and insufficient grounds for declaring permanent cognitive decline. The signal still deserves a board’s attention. Critical thinking, recall, synthesis, and judgment are practiced capabilities. Any organization that routinely outsources the practice should at least wonder what condition those capabilities will be in when the model fails, the context changes, or the decision becomes too important to autocomplete.
A CFO would never let a factory stop testing its backup generator because the electricity had worked perfectly since Tuesday. Leadership teams appear considerably more relaxed about the backup generator between their ears.
These four risks rarely have clean owners. Commoditization sits somewhere between strategy and marketing. Dependency wanders between HR, technology, and occupational health. Slop belongs to everybody, which usually means it belongs to nobody. Atrophy has yet to receive a committee.
The standard risk register remains busy preventing the model from doing something stupid. Good. It should also consider the possibility of the company becoming generic, dependent, unread, and cognitively softer while every dashboard remains green. Put that on slide 17, use the expensive font.
The expensive business of standing still
Corporate caution has developed an impressive wardrobe. It arrives as a governance review, a data-readiness assessment, a regulatory scan, a vendor comparison, an operating-model workshop, and a request to revisit the business case after the summer. Each exercise can be useful once. By the fourth appearance, it has become furniture.
One company waits for the regulator to clarify everything. Another waits for its data to achieve spiritual purity. A third wants the next model release, because committing to technology that may improve later would apparently be unprecedented. Everybody is waiting for a peer to move first, publicly and successfully, so they can follow with the courage of people protected by a benchmark (we all know that big 4 stamped benchmarks save asses, no?) .
This is how paralysis passes for prudence. The organization moves through technological upheaval as an aristocrat moves through a burning house: slowly, with dignity, while insisting the smoke is being assessed.
Visible inactivity looks bad in the annual report, of course, so paralysis usually hires theater as an accomplice. Executives tour Palo Alto. The company holds an AI Day with a futurist, two beanbags, and an alarming quantity of purple lighting. Somebody launches an internal newsletter about the “AI journey.” Attendance at a webinar is recorded as adoption. A pilot that saves eleven minutes in procurement receives a logo and three slides at the leadership off-site.
The company is moving. Nobody can say where. The confetti is now generated faster.
Useful progress is less photogenic. It begins with real bets, named owners, explicit boundaries, and outcomes that customers or employees might actually notice. Failure needs permission before the experiment begins, because retroactive permission is usually called blame. When something fails, the result should be discussed in the same room, with the same volume, same enthousiasm as the announcement that launched it.
Move. Try. Ponder. Break things on purpose. Learn. Move. Repeat. Curse.
The discipline comes from the design of the bet. Keep the blast radius survivable. Put the experiment close enough to the business that people cannot dismiss it as an IT hobby. Run another one before the lessons from the first have been converted into a sixty-page methodology. Above all, resist the ancient corporate instinct to call every mediocre result a successful learning experience and promote its sponsor.
Jamie Dimon offers a useful example precisely because nobody has ever mistaken him for a Palo Alto mystic. JPMorganChase has set a 2026 technology budget of approximately $19.8 billion. That figure covers technology as a whole, with AI, data, cloud, cybersecurity, and infrastructure sitting inside it. In the same annual report, Dimon wrote that AI would affect virtually every function, application, and process in the bank. He also admitted that “we cannot predict the ultimate winners and losers in AI-related industries.”
There it is. A technology commitment approaching $20 billion, accompanied by a public admission that the future remains stubbornly unavailable in spreadsheet format. Dimon named the uncertainty and funded the direction anyway. Compare that with the mid-cap executive committee that spends nine months choosing between two vendor demonstrations, then requests an external benchmark to determine whether it has been sufficiently brave.
Bosch has taken the quieter, more German route. Less keynote fog, more curriculum.
Its AI Academy has already trained roughly 100,000 associates through programs ranging from basic instruction to intensive expert development. The expert track lasts eighteen months. Participants learn data science and engineering while applying those skills to an actual problem from their part of the business. Bosch estimates the investment at a mid-five-figure amount per participant.
That is serious money for training, which explains why many companies would prefer a learning portal and a reassuring email from HR.
Bosch’s internal analysis found that projects delivered by graduates of the expert program generated ten times the original training cost in value within five years. One participant, formerly a controller, used the program to become a data scientist and build a system that predicts supply-chain delivery reliability. The capability stayed inside the company. So did the person who understood the problem in the first place.
Now compare that commitment with the standard multinational AI enablement package: four hours of asynchronous e-learning, an optional Teams call, six multiple-choice questions, and a PDF certificate suitable for hanging beside the GDPR refresher. The course is completed. The compliance box is green. Nobody has built anything.
The cheaper program will look excellent until somebody calculates the cost of having trained an entire workforce to click Next. Boards will never receive certainty about AI. They can insist on disciplined bets, protected learning time, clear accountability, and an honest record of what failed. Those are tangible decisions. Waiting for the future to become obvious is also a decision, although it tends to be recorded later under market disruption.
The most credible sentence a CEO may bring into this discussion is “I don’t know.” It becomes leadership when the next sentence contains a balanced and brave decision.
Above the line, below the line
There is a useful version of the C-suiter-with-AI story. It begins with the calendar.
Below the line sits the administrative sediment of executive life: report preparation, status reviews, meeting summaries, briefing notes, recurring analyses, agenda archaeology, and the weekly search for whoever currently owns slide 19. Let AI absorb as much of that sludge as it safely can. Reclaim the hours. Clear the desk.
Then watch what the C-suiter does with the time.
Above the line sits the work that eventually determines whether the company has a future: customers, products, markets, people, alliances, difficult tradeoffs, and the shape of the business five years from now. The strategic value of AI appears when reclaimed time moves upward. Fewer hours discussing the business. More hours imagining, testing, and building it.
Many executives perform the first half beautifully and forget the second.
Their inbox becomes cleaner while the calendar remains untouched. AI drafts their emails, condenses the reports they were expected to read, prepares talking points for meetings they barely understand, and produces reassuring summaries of decisions they have not yet made. The minutes saved are quietly reinvested in more meetings about saving minutes.
Each shortcut looks reasonable on its own. Together, they create an executive who is impeccably briefed on material nobody absorbed, speaking in prose nobody authored, and approving recommendations nobody seriously challenged.
A Confluent-commissioned survey of 200 UK business owners and C-suite leaders published in 2026 offers a rather spectacular glimpse of where this can lead. According to the detailed results reported by The Register, 62 percent said they used AI for the majority of their decisions. Seventy percent second-guessed their own judgment when it conflicted with the machine’s recommendation. Forty-six percent relied more heavily on AI than on advice from colleagues. Sixty-five percent felt that decision-making had become less collaborative since AI arrived.
This was a small, self-reported, vendor-sponsored survey. It deserves caution, a raised eyebrow, and perhaps a follow-up study conducted by someone who does not sell real-time data infrastructure.
It also deserves a very hard stare.
Taken literally, the results suggest that a substantial number of executives are already handing judgment to systems they cannot fully explain. Allow for exaggeration and the story remains peculiar. Apparently, some leaders believe that publicly admitting they trust a language model more than the people they hired makes them sound modern.
The 70 percent figure may be the most revealing. A model disagrees, and the human begins doubting himself. The machine has no accountability, no career at risk, no memory of the customer meeting in Lyon, and no obligation to explain itself to the board next Thursday. It does, however, respond immediately in confident paragraphs. Confidence has always traveled well in executive circles, even when it arrives without luggage.
A colleague brings context, inconvenient experience, and the possibility of disagreement. The model brings an answer before the coffee cools. Guess who gets described as objective.
AI should sit deep inside the operating system of the company. It can monitor signals, expose patterns, simulate consequences, challenge assumptions, and carry operational decisions at a speed no management team could match. The CEO cannot become its most expensive peripheral.
That distinction grows more urgent as autonomous decision-making spreads. IBM’s 2026 CEO Study, based on research with more than 2,000 chief executives globally, found that AI is already making 25 percent of operational decisions without human intervention. CEOs expect that share to reach 48 percent by 2030.
IBM chose the phrase “rewire the C-suite.” It is the correct verb. Rewiring involves opening the walls, tracing circuits, moving connections, and accepting that the building may be without power for a while.
Most executive committees are applying fresh wallpaper to the fuse box.
They place a chatbot on top of an old process. They appoint a Chief AI Officer without moving decision rights. They create an AI steering committee and leave every functional incentive intact. The organization chart survives, the workflows survive, the meetings survive, and the AI is invited to make all three slightly faster.
Real rewiring changes where decisions happen, how information reaches them, who can intervene, what requires human escalation, and whose name remains attached to the outcome. IBM found that CEOs actively redesigning cross-functional collaboration were more than twice as likely to have delivered on their stated business objectives. That result has considerably more strategic weight than the number of Copilot licenses activated before Christmas.
Brian Solis offers a useful crowbar for this work: WWAID, “What would AI do?”
The careless executive hears permission to outsource judgment. Solis is asking something more demanding. Remove the inherited assumptions for a moment. Take the process apart before improving it. Examine whether the task deserves to survive. Imagine the customer experience, product, or workflow with intelligence present from the first sketch instead of glued on after procurement has signed the contract.
Then use AI to challenge the boundaries of the problem. Ask it to think like the customer who is leaving, the competitor with no legacy systems, the regulator arriving three years early, or the activist investor who has finally read the appendices. Let it surface contradictions, alternatives, and consequences that the room might otherwise avoid. After that, decide.
Use your own voice. Put your own name on it. Explain it to the people who will live with the consequences. Return to the decision when the evidence changes.
AI can illuminate the roads from Alice’s fork. It can map them, model them, price them, and produce an elegant comparison before lunch. We still have to say where the company is going.
Where to go, if you actually care where you go
If the Cheshire Cat were invited to review your AI roadmap, he would ignore the maturity score, the vendor architecture, and the slide showing six workstreams converging heroically toward Q4.
He would ask six questions, each capable of ending the meeting before lunch.
What should now be possible in this company that was impossible before?
The answer must describe an outcome that a customer, employee, patient, citizen, or regulator can see. A customer promise that suddenly becomes affordable. A decision that can happen continuously instead of quarterly. A product personalized to one individual without destroying the margin. A risk detected while something can still be done about it. A market that becomes reachable with eight people where eighty were previously required.
“Improve efficiency” is fog wearing a tie. Put the capability into a sentence that belongs on the front page of the annual report, then see whether the CEO is willing to put it there.
Do we have the nerve and the building blocks to let AI change the company itself?
This is the question missing from most AI strategies. The slides discuss use cases, licenses, training, governance, and approved model lists. All useful. None of them tells you whether the leadership team has the stomach to change how the company actually operates.
The serious version places AI inside the operating system of the business. It changes how information moves, where decisions happen, which workflows survive, how authority is distributed, what customers experience, and what the organization can produce. Eventually, it may challenge the structure of the executive committee itself, which tends to be the precise moment when everybody develops a passionate interest in responsible pacing.
Ambition needs foundations: governed data, usable architecture, secure access, integration, talent, budget, and clear ownership. Yet foundations can become a magnificent hiding place. Some companies have spent three years preparing the runway, repainting the runway, governing the runway, and conducting an employee sentiment survey about the runway.
Nobody has authorized takeoff. The board should ask whether the company is technically capable of reinvention and whether its leaders are personally willing to endure what reinvention does to reporting lines, budgets, territories, and cherished pieces of the organization chart.
Where is AI extending human capability, and where will it remove work or roles?
Say it plainly. Identify whose judgment gains reach, which tasks disappear, which positions change, and what will happen to the people affected. Euphemisms such as optimization, capacity release, and workforce shaping are how responsibility escapes through the ventilation system.
Moderna built tools that expand what its scientists and clinical teams can investigate. Novo Nordisk placed human oversight, workforce upskilling, and domain expertise inside the design of its program. Duolingo initially rewarded visible AI usage, watched the incentive bend behavior, and removed the performance metric.
Each choice reveals what leadership believes people are for. Drift reveals that leadership has avoided choosing.
What are we doing with the time AI gives back?
Ask to see the calendars.
If a serious AI program has been running for six months and the executive calendar remains identical, the company has purchased an expensive way to conduct yesterday’s meetings. The Monday review still lasts ninety minutes. The monthly reporting ritual still consumes three days. Twelve people still attend a status meeting so that two can exchange information while ten contemplate the ceiling.
Where did the reclaimed hours go? Into customers, products, experiments, coaching, difficult decisions, and new markets? Into further headcount reduction? Or did they simply disappear beneath a larger volume of email, generated more efficiently by everyone involved?
Time saved is an operational metric. Time reinvested reveals the strategy.
Which products, business lines, processes, and organizational structures have lost their right to exist?
Nobody enjoys this question. It turns an AI off-site from a pleasant discussion about possibilities into a conversation about consequences.
An AI-native version of the company will carry less historical furniture. Some products were designed around information scarcity that no longer exists. Some management layers exist mainly to collect, summarize, and forward information. Some processes survive because everybody inherited them and nobody owns enough of the inconvenience to kill them. Some business lines will become structurally unattractive long before the annual impairment review discovers the news.
Put them on the table. A company that only uses AI to preserve every existing product, process, hierarchy, and meeting has made a clear strategic choice. It has chosen to automate its attachment to the past.
Can the board see the whole system?
Where is AI running in this company? Which decisions does it influence or make? What data crosses its boundaries? Which models and vendors sit inside the chain? Where can a human intervene? Who monitors performance after deployment? Who owns the outcome when the system fails quietly, confidently, and at scale?
Then add the uncomfortable governance question: how are the directors themselves using AI to prepare for meetings, interpret board papers, challenge management, and reach conclusions?
The answers should exist in plain English. They should not require the CIO to perform a twenty-minute rescue operation involving an architecture diagram last updated by a consultant who has since changed firms.
AIQ becomes visible in the quality of these answers. It shows whether the company has selected a destination, understood the machinery required to reach it, and accepted the consequences of moving. A roadmap that cannot survive these questions has no destination. It is a travel brochure commissioned by people who refuse to leave the hotel.
The Cat will send you back to the fork. The CFO’s pen comes apart again.
AIQ, quoi
Efficiency can pay for the journey. It cannot, (will never) (should never) choose the destination.
AIQ is the capacity of a leadership team to decide what intelligence should make possible, then change the company so that possibility can become real. You see it in the questions the board asks, the assumptions the executive committee is willing to kill, the experiments it funds, the failures it admits, and the hours it deliberately reinvests in customers, products, people, and markets.
AIQ is courage with architecture underneath it and a named human still accountable when the model misbehaves.
The technology can be purchased. Models can be licensed, agents configured, platforms integrated, data cleaned, and several hundred employees certified before the end of the quarter. Vendors will happily place the entire arrangement inside a reassuring circle labelled Enterprise AI Transformation.
The package does not include nerve.
There is no enterprise SKU for curiosity, judgment, or the willingness to redraw a profitable company before somebody else does it for you. Procurement will be disappointed. AIQ cannot be bought for €30 per seat, delegated to the CIO, or parked inside the new Chief AI Officer’s objectives while the rest of the executive committee returns to normal business.
The company either develops it across its leadership system or continues buying increasingly sophisticated tools for avoiding increasingly obvious decisions. The conversation moves beyond tools and begins touching the actual company. Products. Power. Headcount. Decision rights. Management layers. The businesses that may have to die so another version can live.
Part of the room leans forward. The rest waits quietly for the “acculturation” slide to see if something is expected from them. Someone requests a benchmark. Someone else asks whether employees feel psychologically safe using prompts. The CFO begins another controlled disassembly of the Montblanc. Somewhere, a PMO creates a workstream.
That split contains the enterprise AI story of 2026.
One part of the room understands that AI may become an operating layer of the company. It will alter how information moves, how decisions are made, how expertise is distributed, and what the business can offer. Those leaders are prepared to make uncomfortable choices while the evidence remains incomplete.
The acculturation-slide faction will request a revised roadmap for the next meeting. In a few years, possibly during an off-site with a theme involving resilience, they will explain that the disruption arrived suddenly, without warning, after only a decade of warnings.
Back in our board meeting, the CEO reaches the final slide. It promises responsible experimentation, cultural readiness, scalable governance, and a phased approach to value realization. Lunch is waiting. The chief of staff confirms that an updated business case will be circulated. Ninety minutes have passed. The company has discussed AI without deciding what kind of company it wants to become.
Alice is still at the fork. The Cat is still in the tree. The P&L is considerably less patient than either.
So I will ask, one more time, in the flattest voice I own:
“Where do you fucking want to get to?” (pardon my French)
