This is Week 23 of the AI Ops Playbook — Week 7 of Round 3 — and it is about what happens when the person who knows how something works stops being available. Most owners believe they have already solved this. There is a folder. It has SOPs in it. Somebody spent a painful quarter writing them and everyone agreed it was important, and the folder has sat there ever since being pointed at whenever the subject comes up. The folder is not nothing. But it is worth being precise about what it holds, because a documented process and a business that survives losing a person are different achievements, and the gap between them is where this week's four templates live. Here is the test. Take the SOP for something that matters, hand it to a capable person who has never done the job, and let them run it unsupervised. They will get through the normal case. Then they will hit something the document does not cover — an exception, an angry customer, a supplier who says no, a number that looks wrong — and at that moment the document stops helping, because a procedure describes what to do when things are as expected, and the entire value of an experienced person is what they do when things are not. That is the honest shape of the problem. Not that nothing is written down. That what is written down is the least valuable layer of what somebody knows. ## What's New This Week **Tribal Knowledge Capture Interview** (Internal Ops, Advanced) — There is someone in your business who is the only person who knows how something works, and you know exactly who and exactly what. The obvious move is to ask them to write it down. It does not work, and the reason it does not work is the interesting part: an expert asked to describe their job produces the idealized happy path, which is already in the SOP, because twenty years of practice turns knowledge into instinct and instinct does not narrate itself. So this template does not ask them to document anything. It generates a custom interview guide built on six elicitation moves that route around the expert's own summary of their work — ask about exceptions rather than procedures, ask for specific remembered incidents rather than generalizations, ask where the written process and the real practice disagree, and ask them to predict what a competent new person will get wrong in the first month. Then a second prompt turns the recording into a structured document whose Exceptions section is longer than its Process section. That ratio is how you know the interview worked. **Role Handoff Doc Builder** (Internal Ops, Intermediate) — Somebody is leaving and the two weeks you have are already half spent. An offboarding checklist will recover the laptop and close the accounts; it will not tell the next person that the biggest client only answers text messages, that the Thursday report is manually corrected before it goes out, or that a half-finished supplier renegotiation expires in April. This template builds the document that carries the job itself, organized around a specific principle: anything with a ticket, a contract, or a calendar invite will survive the transition on its own, because a system is holding it. What disappears is everything held only in one head. So the document is weighted toward in-flight work whose true state differs from its reported state, relationships that run through an individual rather than the company, unwritten agreements, and landmines — the things that look wrong but are intentional. It includes the section people leave blank out of politeness and should not: what I would fix if I were staying. **Team Training Module Builder** (Internal Ops, Intermediate) — You wrote the SOP, you sent it to the new hire, they read it, they said it made sense, and then they did the job wrong anyway. That is not a discipline problem and it is not a bad document. A reference and a teaching document are different objects, and most small businesses only ever write the first. This template converts what you already have into a module built to produce capability rather than familiarity: the reason the task matters stated concretely, the concepts taught before the steps, one narrated worked example, three practice cases where the third has no clean answer, an answer key that treats the tempting wrong answer as seriously as the right one, and a competence check the learner can genuinely fail. It also audits your source document for the curse of knowledge and hands back a list of every place the SOP assumed something it never explained. **Meeting-to-Decision Log** (Internal Ops, Beginner) — Every business has a policy nobody can explain. Somebody set it for a good reason, the reason was never recorded, and now the team either works around it forever or quietly reverses it and rediscovers the original problem the hard way. This template reads meeting notes and extracts the decisions — each with the alternatives that were rejected and why, the assumptions it rests on, whether it is reversible, and a concrete trigger for revisiting it. It also produces three sections a normal meeting summary never does: the decisions that took effect by default because a deadline passed and nobody chose, the disagreements that were never resolved, and an honest list of every place your notes recorded an outcome without a reason. It is instructed never to invent a rationale, because a plausible reconstructed reason is worse than an admitted gap. ## Why These Four Together — Four Layers, Not One The reason these ship as a set is that they are not four versions of the same idea. They are four distinct things that disappear, and a business can lose any one of them while holding the other three. The first is **what somebody knows but has never articulated**. This is the layer under the SOP: the exceptions, the thresholds, the early signals that a job is going to be a problem. It cannot be recovered by asking, only by interviewing, which is why the lead template is an interview protocol rather than a documentation prompt. The second is **what somebody is currently carrying**. Not knowledge at all, but state — the half-finished negotiation, the promise made on a call, the client who is one bad week from leaving. This layer has a deadline attached to it. It is worth the most on the day someone gives notice and is worth nothing a month after they have gone. The third is **the ability to make someone else good at it**. A business can have excellent documentation and still take nine months to bring a new person up, because documentation is not teaching, and the difference is practice, feedback, and a bar that can be failed. And the fourth is **why anything is the way it is** — which brings us to the part of this week worth sitting with. The first three are all about a person leaving. The fourth is not. Decision amnesia happens to teams that are completely intact. Nobody resigns, nobody retires, nobody goes on leave, and eighteen months later the same five people are in a room relitigating something they themselves settled, because what got written down was the outcome and not the reasoning. This is the most common continuity failure in small business and the one nobody classifies as a continuity failure at all, because there is no departure to blame it on. It just presents as meetings that feel repetitive and rules that nobody quite trusts. That is also why the decision log is the beginner-level template of the four and the one to start with this week. It requires no interview, no scheduling, and no departure. You can do it tonight with the notes from a meeting you already had. ## The Part That Makes This Hard Knowledge capture has no deadline until it has an emergency. It is permanently the thing you will do next quarter, and it stays there because nothing breaks while you do not do it — right up until somebody gives two weeks' notice. And two weeks is precisely the worst possible moment: the person is already mentally gone, you are panicking, and the interview you should have run calmly a year ago now has to happen between exit paperwork and a leaving lunch. So here is the recommendation, and it is the whole point of the week. Run the tribal knowledge interview on someone who is not leaving. Someone with no notice period, no resentment, and no distraction, who will give you a better hour than any departing employee ever has. It is cheap insurance now and it is unbuyable later, and the only reason it does not happen is that there is never a Tuesday when it is urgent. Pick the person you would least like to lose. That is the whole selection process. ## A Note on Cadence — Twelve Wednesdays Straight This drop lands on its regular **Wednesday** for the twelfth consecutive week, an unbroken run from Week 12 through Week 23. The library is now at 137 templates. Next week is Week 24 — the Round 3 finale, and a trending-AI week. ## Get Started All four templates are available now for Pro members: - **Tribal Knowledge Capture Interview** — [open the template →](/templates/tribal-knowledge-capture-interview) - **Role Handoff Doc Builder** — [open the template →](/templates/role-handoff-doc-builder) - **Team Training Module Builder** — [open the template →](/templates/team-training-module-builder) - **Meeting-to-Decision Log** — [open the template →](/templates/meeting-to-decision-log) If someone is leaving right now, start with the handoff builder and do not wait until next week. If nobody is leaving, that is the best possible time to use the other three, and it is the only time they are easy. The one habit worth taking from all four: when you write down what was decided, write down what you rejected and why. It costs an extra two minutes in the moment and it is the single cheapest thing you can do to stop your business from having to learn the same lesson twice.