AI Academy · Book
Executives & Directors · Module 10 · Chapter 009

Module 10 Synthesis — The Future AI Leader

Nobody can say where AI will be in five years, and the future AI leader does not need to. The job is to keep the organization able to benefit whichever way the technology moves: watch for thresholds, test on your own work, commit at the speed a decision can be undone, and review on a date. Ten modules reduce to that one habit.

≈ 13 min read

After this chapter you can

  • Recall the idea to keep from each Module 10 chapter and each module of the course, without re-teaching them.
  • Explain why a posture beats a prediction when AI capability, cost and rules keep moving.
  • Run one leadership loop - watch thresholds, test on your own work, commit at the speed of undo, review on a date.
  • Separate what will keep changing from the principles that will not, and plan around the second.
  • Name the parts of the job an executive cannot delegate, including AI literacy and board reporting.

In January 2026, PwC published its annual survey of chief executives: 4,454 of them, in 95 countries and territories. Asked what AI had done for their companies over the previous twelve months, 30 percent reported higher revenue and 26 percent lower costs. Fifty-six percent reported neither. Only 12 percent reported both, according to PwC’s release as reported in the press12. Each figure is a share of all the CEOs surveyed, so the groups overlap: the 12 percent who saw both are counted inside the 30 and the 26.

In a survey of 4,454 CEOs, 56 percent reported no revenue or cost benefit from AI in a year, and 12 percent reported both.56%of CEOsNo revenue gain and no cost saving from AI in12 months12%of CEOsBoth higher revenue and lower costsSource: PwC, 29th Global CEO Survey · January 2026
Figure 10.9.1 The models were available to everyone. The difference was the organization they arrived in.

These are self-reported figures, and twelve months is a short window. But look at what set the 12 percent apart. It was not access to better models. As The Evolution of AI Models showed, the leading models now sit close together and anyone can rent them. PwC found that the companies seeing both gains had stronger foundations in place: defined roadmaps, and technology environments built for integration1. The difference lay in the organization the AI arrived in, and in the leaders who had prepared it.

That leaves one question to close the course. If the technology keeps changing faster than any plan, what kind of leader keeps an organization among the 12 percent?

Posture, not prediction

The future AI leader is not the person with the best forecast. Forecasts move too fast to lead by. The researchers surveyed for Preparing for AI Uncertainty and Strategic Change moved their own estimate of when machines would beat people at every task by 13 years in a single year3. A leadership team that bets the company on one forecast inherits its instability.

The future AI leader prepares the organization rather than the forecast, through a steady rhythm of watching, testing, committing and reviewing.Prepare the organization, notthe forecast.Watch, test, commit, review - on a date.
Figure 10.9.2 The future AI leader is judged by how well the organization adapts, not by how well the leader predicted.

What replaces the forecast is a posture: a steady rhythm of watching, testing, committing and reviewing that keeps the organization able to create value whichever way AI moves. It asks for judgment more than expertise. An executive does not need to build a model. An executive does need to decide which problems matter, how much evidence is enough before money and people move, who owns the outcome, and when stopping is the right result. Each module of this course supplied part of that judgment. Read together, they are one job.

The course in one page

Not one of the ideas in this last module depends on knowing which model or vendor will lead next year. Each replaces a prediction with something a leadership team can check for itself: a threshold crossed on its own work, a door it chose for an agent, a cost per completed task, a market it mapped, a decision it can undo. The same is true of the course as a whole. Its ten modules can be read as ten questions that a leadership team asks of any AI opportunity, in roughly the order it should ask them.

One line per module of the course, from redesigning work in Module 01 to preparing rather than predicting in Module 10.ModuleThe idea to keep01 The AI RevolutionValue comes from redesigning work, not from adopting tools02 Understanding AIFluent is not verified: know what it does, how reliably, and how it fails03 Business ValueName the outcome, prove it, then capture it04 AI StrategyA coherent set of choices, not a list of projects05 Enterprise Use CasesValue concentrates in a few workflows; analyze is not authorize06 AI RisksWhat could go wrong, what controls exist, who accepts the rest07 AI GovernanceOne connected system that makes responsible AI faster to scale08 AI EconomicsJudge the cost of each unit of value, at ten times the use09 Executive RoadmapOnly the quarter is a contract; stage money behind evidence10 Future of AIPrepare, do not predict
Figure 10.9.3 Ten modules, ten questions: why it matters, what it is, what it is worth, what to choose, where, what can go wrong, who decides, what it costs, in what order, and what next.

Three threads run through every row. The first is redesign: from the electric motors of Module 01 to the AI-native workflows of this module, value has come from changing the work, not from installing the tool. The second is evidence before scale: a pilot proves possibility, production proves the organization can run it, and scale proves the result repeats at a sensible cost, as From AI Pilot to Production to Scale set out. The third is ownership: every risk, every control and every outcome needs a named person who accepts it. A leader who holds those three threads can take on a technology the course never mentioned and still ask the right questions.

One loop for a moving frontier

AI Leaders vs AI Followers, early in the course, described the behaviors that separate leaders on any single initiative: choose, own, change and prove. The loop below sits one level up. It sets the rhythm at which the whole portfolio responds to a frontier that will not stand still.

The future AI leader's loop - watch for thresholds, test on your own work, commit at the speed a decision can be undone, and review on a fixed date.WatchThresholds, notheadlinesTestOn your own workCommitAt the speedof undoReviewOn a date: keep,change, stopLeader's rhythmEvery quarter
Figure 10.9.4 One loop, run on a fixed rhythm. The review is the step most easily skipped: launch, then never decide again.

Watch for thresholds, not headlines. A release, a price cut or a new law is a signal, not an instruction. The useful question is whether something your plan depends on has crossed a level that matters: an error rate on your own cases, a cost per completed task, a date in a market where you sell. Give each threshold an owner who reports on it, so that watching is a duty and not a mood.

Test on your own work. Public benchmarks saturate and vendors’ demonstrations are chosen to impress. A small evaluation set built from your own cases, run whenever something changes, tells you more than any league table. Keep tests cheap and bounded, with a stop rule written before they start.

Commit at the speed the decision can be undone. No-regret moves can go now; options buy the right to act later; big bets wait for strong evidence. Before a commitment that is hard to reverse, run a pre-mortem: imagine it is two years later and the initiative has failed, then write down why. In a classic experiment, people asked to explain an outcome as if it had already happened produced about 30 percent more reasons than people asked to explain one that only might happen4. Gary Klein turned that finding into a short meeting any project team can hold5. Where the commitment hands work to an agent, the controls of Module 06 still apply: least privilege, approval and reversibility.

Review on a date. Every commitment carries the date on which it will be decided again, and three possible answers: keep, change or stop. A stop is a result, not a failure, because it frees people and money for something better. It is also the easiest step to skip: launch, report activity, and never decide again.

What will change, and what will not

Almost everything specific in this course will date. Some of it has dated while being written. That is not a reason to discount it. The specifics illustrate principles that have outlasted several technologies already.

Models, costs, autonomy, rules and supply will keep changing; the need for redesign, evidence, owners, unit economics and care for people will not.Will keep changingWhich model leadsWhat a task costsHow long an agent can work aloneThe rules in each marketWho can supply chips and modelsWill not changeValue needs redesigned workEvidence comes before scaleEvery outcome needs an ownerCost is judged per unit of valuePeople carry the transition
Figure 10.9.5 Plan around the right-hand column, and watch the left-hand one for thresholds.

The left-hand column is where the loop watches and tests. The right-hand column is what the loop protects. When a proposal arrives that depends on the left-hand column holding still, such as a five-year contract priced on today’s model or a design that assumes one supplier, treat it as a one-way door and slow down. When a proposal ignores the right-hand column, such as a rollout with no owner or a business case with no cost per outcome, the technology is not the problem.

The leader’s own part

Some parts of this job cannot be delegated. The first is literacy. An executive who has never used the tools on their own work cannot judge the claims made about them, and will be either too easily impressed or too easily frightened. The second is the demand for evidence: asking, at each stage, what has been proved and what is still assumed. The third is protection for stopping. People stop failing work only when stopping is rewarded, and that signal can only come from the top. The fourth is reporting: making sure the board hears about AI risk and value through a channel that exists, not by accident.

Story: the forecast that ordered the wrong shoes

One well-documented post-mortem of a consumer-goods maker handing its decisions to an algorithm comes from before generative AI, which is part of why it is useful. Nothing in it depends on the technology of the day.

In 1999 Nike began installing a new demand-planning system, part of a supply-chain and enterprise-software program that was budgeted at about 400 million dollars. The planner was meant to forecast demand and drive orders to the factories that made Nike’s shoes. It was installed in a hurry, heavily customized, and run while the old systems were still in place. Planners were not well trained before it went live. Data had to be downloaded and reloaded by hand, sometimes weekly, to move between systems. And the system’s orders went straight to factories, where they became shoes: thousands too many of a slow-selling model, and thousands too few Air Jordans7.

Nike's demand-planning system was rushed in from 1999, sent wrong orders to factories, forced a February 2001 profit warning, was withdrawn from core use, and the program was then rebuilt with training and staged rollouts.1999Rushed installHeavycustomization, littletraining1999-2000Orders gowrongWrong shoes reachthe factoriesFeb 2001DisclosureForecast cut; sharesfall 20%Spring 2001Pull backPlanner withdrawnfrom core use2001-2002Gated rolloutTrained, staged,timed forquiet weeks
Figure 10.9.6 The algorithm was not the only failure. The commitment was: its output went straight into a decision that could not be undone.

On 26 February 2001, Nike cut its forecast for the quarter to 34 to 38 cents a share, from 50 to 55 cents. It blamed complications from implementing its new demand and supply planning systems: product shortages, excess inventory and late deliveries8. The cost was put at about 100 million dollars in lost sales, and the share price fell 20 percent. Phil Knight, the chairman, summed it up: “This is what you get for 400 million, huh?”7

Nike lost about 100 million dollars in sales and 20 percent of its share price, then trained each customer service user for 140 to 180 hours before access in later rollouts.100MLost sales (dollars)Estimated cost of the failure20%Share price fallAfter the February 2001 warning140-180 hTraining per userIn the rollouts that followed,before any accessSource: CIO, Nike rebounds · 2001-2004
Figure 10.9.7 The recovery cost more in time and training. It cost far less than the failure.

What Nike did next is the more instructive half. In the spring of 2001 it stopped using the demand planner for its short- and medium-range sneaker planning and moved that work into its main enterprise system. In the rollouts that followed, it agreed one global process template before configuring anything, gave each US customer service representative 140 to 180 hours of training, locked people out of the system until they had completed it, and timed each go-live for a quiet holiday weekend. The program’s cost rose to about 500 million dollars. The company later claimed its manufacturing lead time had fallen from nine months to six, though an analyst quoted at the time disputed the gain. Its chief information officer drew the lesson in five words: “You can never train enough”7.

Read it against the loop. No one was watching a threshold: the planner’s errors surfaced publicly, in a profit warning. The system was not tested on real orders at real scale before its output was trusted; the CIO later said the rollout could probably have taken more time. The commitment was a one-way door: a factory order becomes shoes, and shoes cannot be un-made. And the review came from the financial results, not from a date the company had set. The recovery was the loop, installed after the fact: staged, gated, trained and timed so that each step could be undone. Generative AI could make this kind of error cheaper to commit and faster to spread, because a model’s output can flow into orders, prices or messages with no one in between. That is a risk to plan for, not yet a pattern documented at scale.

What this means for leaders

The future AI leader holds two things at once: ambition about what AI can do for the organization, and discipline about how fast any single commitment is allowed to become irreversible. Many of the failures in this course’s cases came from holding only one. Fear without ambition produces pilots that never scale and a strategy that waits for certainty. Ambition without discipline produces 2001 at Nike: a confident system wired straight into decisions no one could undo.

The practical answer is not a new framework. It is the loop, run on a calendar, by people who own its parts, with the three threads of redesign, evidence and ownership holding it together.

Check yourself

  1. The future AI leader’s main job is to predict which AI technology will win.
  2. In PwC’s January 2026 survey, more than half of CEOs reported no revenue gain and no cost saving from AI.
  3. An executive must be able to build AI models to lead AI well.
  4. Stopping an AI initiative at review is a sign the loop has failed.
  5. Nike’s 2001 losses came from trusting a new planning system’s output before it had been proved at scale.
  6. Imagining that a project has already failed helps people find more reasons it might fail.

Reflection: your first review date

What comes next

This closes the content of Level 1. The work itself, though, will be carried out by the people who run your AI teams day to day. Level 2, Engineering Managers, is written for them: how to turn the choices made here into teams, delivery and operations that hold up.

Laws referenced

Not legal advice. Laws change; verify before relying on this, and consult counsel for decisions.

EU AI Act · EU

Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744

Risk-based rules. Prohibited practices include social scoring, untargeted scraping of facial images, and emotion recognition in workplaces and schools (with narrow exceptions). High-risk systems (Annex III: biometrics, safety components of critical infrastructure such as energy, water and traffic, employment and worker management, credit, education, essential services, law enforcement, migration, justice) need risk management, data governance, documentation, logging, human oversight, human oversight that keeps people able to understand the system, notice automation bias (over-reliance on its output), override it or stop it (Art. 14(4)), appropriate accuracy, robustness and cybersecurity (Art. 15), automatic logging of events (Art. 12), a provider quality-management system (Art. 17) and conformity assessment. An Annex III system is not high-risk if it poses no significant risk of harm, for example a narrow procedural or preparatory task that does not replace human assessment; systems that profile people are always high-risk, and a provider relying on this exception must document it and register (Art. 6(3)). Deployers of high-risk AI must use it as instructed, assign competent human oversight, monitor its operation, keep logs for at least six months and report serious incidents (Art. 26); employers must inform workers' representatives (Art. 26(7)). Public bodies, private providers of public services, and deployers of credit-scoring or life and health insurance pricing systems must carry out a fundamental-rights impact assessment before first use (Art. 27). Providers must run post-market monitoring (Art. 72). A deployer that puts its name on a high-risk system, substantially modifies it, or changes its purpose so that it becomes high-risk takes on the provider's obligations (Art. 25(1)). A substantial modification (Art. 3(23)) of a high-risk system needs a new conformity assessment, unless the change was pre-determined and documented at the first assessment, as with planned continuous learning (Art. 43(4)). Providers of general-purpose AI models (from 2 Aug 2025) must keep technical documentation, have a policy to comply with EU copyright law including text-and-data-mining opt-outs, and publish a sufficiently detailed summary of training content (Art. 53). Research, testing and development before a system is placed on the market or put into service is outside the Act, except testing in real-world conditions (Art. 2(8)). Since the 2026 Omnibus, the Art. 4 AI-literacy duty is an obligation of effort (take measures to support literacy), not of result. Fines reach EUR 35 million or 7% of global turnover for prohibited practices.

  • 2024-08-01 — Entered into force
  • 2025-02-02 — Prohibited practices (Art. 5) and the AI-literacy duty (Art. 4) apply
  • 2026-07-27 — Omnibus softens Art. 4: providers and deployers must take measures to support AI literacy; no specific level must be guaranteed
  • 2025-08-02 — General-purpose AI model obligations apply; governance and penalties regime in place
  • 2026-08-02 — Transparency duties (Art. 50) apply: disclose AI interaction, label synthetic and deepfake content (marking for generative systems already on the market: 2 Dec 2026)
  • 2027-12-02 — High-risk obligations for Annex III systems (e.g. hiring, credit, education, essential services) - moved from 2 Aug 2026 by the 2026 Omnibus
  • 2028-08-02 — High-risk obligations for AI in products regulated under Annex I

Last verified 2026-10-06 · official text

EU Digital Omnibus on AI · EU

Regulation (EU) 2026/1744

First amendment to the AI Act. Defers high-risk obligations (Annex III to 2 Dec 2027, Annex I to 2 Aug 2028), adds two prohibited categories, softens the Art. 4 AI-literacy duty to "take measures to support", and simplifies some compliance duties. Art. 50 transparency duties still apply from 2 Aug 2026, with one transition (new Art. 111(4)): providers of generative AI systems placed on the market before 2 Aug 2026 must meet the Art. 50(2) marking duty by 2 Dec 2026.

  • 2026-07-24 — Published in the Official Journal
  • 2026-07-27 — Entered into force
  • 2026-12-02 — Grace period ends for safeguards against two new prohibited uses (non-consensual intimate imagery, child sexual abuse material)
  • 2026-12-02 — Art. 50(2) marking duty applies to generative AI systems placed on the market before 2 Aug 2026 (Art. 111(4))

Last verified 2026-10-10 · official text

Directors' duty of oversight (Delaware "Caremark" doctrine) · US - Delaware (case law for most large US companies)

In re Caremark International Inc. Derivative Litigation (Del. Ch. 1996); Marchand v. Barnhill (Del. 2019)

Directors breach their duty of loyalty if they make no good-faith effort to put a reporting system in place for mission-critical compliance risks, or ignore red flags it raises. Where AI is mission-critical, a board that has no way to hear about AI risk is exposed. Liability is hard to establish, but the duty shapes what boards should ask to see.

  • 1996-09-25 — Caremark decision
  • 2019-06-18 — Marchand v. Barnhill: board must make a good-faith effort to oversee mission-critical risks

Last verified 2026-10-08

References

  1. PwC. Leading through uncertainty in the age of AI: PwC's 29th Global CEO Survey. PwC. 2026.
  2. IT Channel Oxygen. Most CEOs say AI yielding no returns: study. IT Channel Oxygen. 2026.
  3. Katja Grace, Harlan Stewart, Julia Fabienne Sandkühler, Stephen Thomas, Ben Weinstein-Raun, Jan Brauner and Richard C. Korzekwa. Thousands of AI Authors on the Future of AI. Journal of Artificial Intelligence Research 84, Article 9. 2025.
  4. Deborah J. Mitchell, J. Edward Russo and Nancy Pennington. Back to the future: Temporal perspective in the explanation of events. Journal of Behavioral Decision Making 2(1), 25-38. 1989.
  5. Gary Klein. Performing a Project Premortem. Harvard Business Review. 2007.
  6. Supreme Court of Delaware. Marchand v. Barnhill, 212 A.3d 805 (Del. 2019). Justia US Law. 2019.
  7. Christopher Koch. Nike rebounds: How (and why) Nike recovered from its supply chain disaster. CIO. 2004.
  8. Marc L. Songini. Nike blames financial snag on supply-chain project. Computerworld. 2001.

Further reading

Sources last verified 2026-10-08.