The distinction, in one sentence
Drift versus design is the difference between an organisation that adopts AI through a thousand small decisions nobody quite made, and one that decides in advance where human judgement has to remain.
That sentence is the whole keynote and the rest is evidence for it. The term was first published on 16 November 2025, in Box of Amazing, and the dated record is kept at drift versus design so anyone can check when it was said rather than take it on trust.
What makes drift hard to see is that nothing goes wrong while it happens. The work still ships. The output often improves. No control fails, no incident is logged, and no meeting is held about it. An organisation that is drifting looks, from the inside, exactly like an organisation that is doing well.
Drift sometimes looks like progress: more output, more happening, and you have handed over the things you were good at without redefining the role.
An engineer described this better than I can. He is not vibe coding. He is careful, he checks what comes back, and he ships more than he ever has. What he said was that he is doing a different job than he was three years ago, that nobody told him the job had changed, and that it feels hollow. Every decision that got him there was a good one. The role description underneath simply never moved, and no dashboard would show you that.
AI is making everyone faster but nobody better. That is the sentence the room usually argues with first, and the argument is the session.
Why this is a judgement problem before it is a jobs problem
AI rarely takes a whole job. It takes the first draft, the initial analysis, the routine review, the first pass over the file. Those tasks were doing a second job nobody costed: they were how people built the judgement that made them senior.
So the loss is deferred and it is invisible at the moment it occurs. A team can lose the work that produces expertise for two years and show better numbers throughout, and then discover that the person who was supposed to be able to check the machine learned the job in a period when the machine did the parts you learn from. The argument is set out at missing rungs and who supervises work they cannot do.
This is why the keynote does not open with adoption statistics. Adoption is the question every other session in the programme is already asking. The question this one asks is narrower and harder: which decisions in this organisation require judgement that somebody here still has, and how would you know if that stopped being true?
What the room is asked to do
The session ends with one exercise and it is deliberately uncomfortable. Take the decisions your organisation makes that would be expensive to get wrong, and sort them into three: the ones you are happy to hand over, the ones where the machine drafts and a person decides, and the ones a person must do from the start.
Most leadership teams find they have never written that list down. Some find they disagree with each other about it, which is more useful than agreeing, because the disagreement is the thing that was going to be settled by drift.
The point is not the list. The point is that a list written on purpose is a design decision, and the absence of one is also a decision, made by nobody, that everything is negotiable.
Where the evidence comes from
Every claim in the talk is traceable and graded, and the grading includes what each source does not prove. That discipline is the reason the argument survives a hostile question from the third row.
Two examples of the kind of thing it rests on. Japan’s labour policy institute surveyed 22,000 employees and found employer AI use at 12.9 per cent, which is a very different picture from the one on most conference stages. And the human-factors literature going back to Parasuraman and Riley in 1997 already described the failure mode: people do not distrust a system that has been reliable, and reliability is exactly what removes the reasons to doubt.
The full base is open at the evidence base, and what the research does not yet know is listed rather than hidden.
Formats, and who it is for
Forty to ninety minutes as a keynote, or a closed board session where the sorting exercise is run live. It works on a main stage and it works better in a room where people can argue. Boards, executive committees, leadership offsites and the opening or closing slot of a corporate conference.
Two other versions of the same argument exist. We Are Superheroes covers the seven human skills that grow more valuable as the tools spread, for a whole organisation rather than a leadership team. WTH, What the Human is a live judgement test run in the room, for boards, December to March only. All three are described at keynotes.
He is the wrong choice for a tools demonstration, a vendor pitch or a technical AI briefing, and for any brief that needs an optimistic message with the difficult part removed. If that is the brief, the guide to choosing an AI keynote speaker names four other kinds of speaker and when to pick each.
