The first week of an enterprise AI program does not look like the keynote. There are no robots and no fog machines. There is a conference room with bad coffee, a stack of printed runbooks, and forty people who want to know whether this thing is going to replace them. It is not. The whole first week is built to make that clear.
Monday — the room, not the model.
We do not open with the model. We open with the people who will use it. The first session is ninety minutes and the AI barely comes up. We talk about what each person does now, what stays theirs, and what the system will take off their plate. Augment, never replace is not a slogan here — it is the literal agenda of day one.
Mon 09:00 · Kickoff — who does what
The runbook stack, still warm from the printer
Laminated cheat sheet · v1Wednesday — the laminated cheat sheet.
By midweek the abstract becomes a card. We laminate a single-page cheat sheet for every role: what the AI can do, what it must never do alone, and the exact phrase to use when you want a human to check its work. People keep these. We find them taped to monitors months later, soft at the corners. A laminated card outlasts a slide deck because you can hold it at 2pm when something feels off.
- What it does. The three tasks the system actually handles for this role. Not a feature list — a job list.
- What it never does alone. The decisions that always route to a person. Written in plain language, no hedging.
- The escalation phrase. One sentence that flags a case for human review. Memorable on purpose.
Week one is not about teaching people to trust the AI. It is about teaching them exactly where not to.
Thursday — runbooks, not vibes.
The runbook is the spine of the week. Every workflow the AI touches has one: the normal path, the edge cases, and the rollback if the output looks wrong. We walk each team through their runbook by breaking it on purpose — feeding the system a bad input and watching the human catch it. The point is not that the AI is perfect. The point is that the catch is rehearsed before it is needed.
Thu 15:40 · A team catches a bad output on purpose. The runbook held.Friday — the quiet metric.
We close the week with one number, and it is not accuracy. It is how many people in the room can say, without checking the card, where the human stays in the loop. On Friday that number is high, and that is the only week-one result we care about. The model will improve over months. The habits set this week are the thing that has to hold.
Nobody lost a job this week. Everybody got a card, a runbook, and a clear line. That is what a Responsible AI rollout looks like from the inside — slower than the demo promised, and far more likely to still be running a year from now.
— Suhail Abidi. Austin, TX. May 2026.
Comments