A closed cycle for a group of executives to stop approving what they can't examine. This isn't a course on artificial intelligence: it's judgment practice, on real decisions, under the new condition AI has created.
One cohort per cycle, from the same organization. Each participant enters with their own diagnostic, and cohort results are analyzed in aggregate.
Almost every organization has already trained its teams on tools. Few have addressed the question that remains afterwards, and it's an executive's question, not an engineer's.
The executive approves reports, analyses, and recommendations they would have no way of producing and no way of checking. The approval still happens, and the name on it is still theirs. What disappeared was the examination.
The question "how far can this process run on its own, and what control has to remain on my side" has no owner in most companies. Each area answers it differently, and no answer is auditable six months later.
Each module is a prerequisite for the next and closes with a deliverable. Two weeks per module in the base cycle, four in the extended cycle. There are four deliverables, one per module, ready to be used in the following week's operation — working instruments, not reading material. They come out finished and applied to the participant's own case, never as a blank template.
Arriving at a committee with a formed position, even with sources that contradict each other and little time. Ceasing to be the person who reads everything and becoming the person who decides what's worth reading.
Leaves with: a script of their own for reaching a formed position in half a day, run on a real decision from their own company.Assessing a proposal and a vendor by their own criteria. What to approve, what to veto — and, above all, how to make the veto stick when others want to approve.
Leaves with: an instrument for assessing proposals and vendors, applied to a decision that's on the table now.Leading people with a real fear of the change, redefining what's expected and what's required. It's the module where the conversation stops being about technology and becomes about people.
Leaves with: the script for the conversation that redefines what's expected of the team, and a redesigned process that names who signs each step.Holding a position of authority outside their own team, backed by evidence, not assertion. It's where the program stops being development and becomes positioning.
Leaves with: the piece they use to hold their own position outside the team, built on evidence from the cycle, not on an opinion about AI.The unit of work isn't the lecture, it's the case. Groups receive a real decision with two positions in conflict, and have to leave the session with a recommendation that survives the rebuttal.
Cases from technology, human resources, and governance: an adoption that promises to prevent incidents, a screening process that decides about people, AI use that nobody authorized, an adoption target that makes the operation more expensive. Some are built for practice; three are public facts, reported by the press.
The house rule: one in every three cases has to end with a decision to grant more autonomy. A method that only brakes isn't a method, it's a temperament.A protocol isn't a document template: it's the sequence of questions that leads from raw information to a defensible decision. Each one is run in the room, on the participant's own case, and reviewed with the group before the module closes.
Each cycle also produces an aggregate cohort report, with the decision pattern that appeared in the group — without exposing individual answers."Not being agentic isn't a failure. It's a choice, when the reach of the error doesn't pay for the autonomy."
Part of the market defines as advanced only what runs without a human at every step, and from there concludes that requiring approval is backwardness. That's a definition of architecture, not a judgment of maturity — but it reaches the board as though it were a grade.
In this program the exercises operate at a declared level of autonomy, with the human inside the loop. And the participant isn't required to agree: faced with a case that calls for more autonomy, the exercise is to argue whether the premise holds there, not to obey it. That's what trains judgment instead of compliance.
In 2026, after two serious incidents, one of the most AI-intensive companies in the world reinstated human approval for assisted code. It stepped down a rung on purpose. The yardstick isn't conservatism: it's the same one that company ended up applying, late.
The content was developed by someone with firsthand executive responsibility, grounded in public, verifiable evidence. Where there's a third-party framework, it's cited; where there's an extension of our own, it's declared as such.
A forty-year career in four global corporations, from engineering to executive leadership, the final years as CTO for Latin America. Fortune 500 operations, teams of hundreds of people, decisions with irreversible effect — which is exactly the subject matter of the program.
The AI adoption scale starts from the public yardstick by Mike Taylor and Every. The reversibility axis and the remaining-control rule are original Prædictor extensions. Governance rests on the NIST AI RMF; the analysis of power and influence, on French and Raven.
Before there was a yardstick for deciding about AI, there was the method with which Prædictor treats executive development. There are five fronts, and the program takes its name from the second. The full account of all five lives on the Career Map page, which is where they operate end to end.
The program is designed for a group from the same organization, or from organizations that share the same kind of decision. It isn't an open course, and there's no shopping cart.
People who already answer for results and have to decide about AI without being the technical person in the room.
Each participant enters with their own position measured on two axes, and cohort-level results are analyzed in aggregate.
Sessions run on a fixed calendar, with group work between them on a case the cohort carries from start to finish.
The organization receives the decision pattern observed in the cohort and where the governance gaps are.
One module every two weeks. Each group carries one case from start to finish and leaves with the four deliverables applied. It's the minimum design in which the method completes a full cycle.
The same design with twice the time per module, and the extra time isn't more content: it's more repetition. More cases from the library, decisions defended individually in front of the group, and application followed up between sessions.
The choice between the two isn't a matter of budget, it's a matter of ambition. Eight weeks build common criteria and leave four instruments running. Sixteen build practice — and practice is what makes the criteria survive the first difficult room. The proposal is put together by scope: how many people, on what calendar, and what kind of decision they sign.
The conversation starts with scope, not with a budget: how many people, what kind of decisions they sign off on, and what the reach of the error is in those decisions.
I want this program at my company