Leaving well is a knowledge problem

The last thing most people do before leaving a role is write a document. A long one. Strategy, status, open items, who to talk to, what’s on fire. It gets shared, everyone says thank you, and about four weeks later nobody can find it.

I’ve written that document before. I’m not proud of it.

The problem isn’t effort. It’s that a handover doc answers the questions you thought to ask during your last week. The questions your successor actually has arrive in month three, and they’re never the ones you anticipated.

Because what walks out the door with you isn’t information. Most of the information is already somewhere – in the CRM, the board deck, the shared drive, a Slack thread from November. What walks out is judgment. Why we killed that campaign after two good months. Which of the two agencies is genuinely good and which one is just fast to reply. Who on the team is quietly ready for more and doesn’t know it yet. Why the number on slide eleven is technically true and practically misleading.

None of that survives a goodbye lunch.

So I built a brain instead

I pointed Claude at a twelve-month scan of my Slack and my inbox. That gave me a skeleton: the decisions, the recurring topics, the threads that kept coming back, the rough shape of a year. Useful, and completely insufficient. The scan knew what happened. It had no idea why.

So I sat down and talked. A verbal braindump, and then I let it interview me to fill the gaps between what the systems showed and what I actually knew.

That interview was the part I didn’t expect. It kept pushing on the seams – a decision that appeared fully formed in March with no visible discussion before it, a vendor who stopped being mentioned in June and never came back, a metric we reported for three quarters and then quietly dropped. Half my answers were things I’d never written down anywhere and would never have thought to include in a memo.

What came out isn’t a document. It’s a small set of versioned markdown files, each one owning a single thing:

• a master doc carrying the narrative and the open decisions
• one for the team – how each person works, what they’re ready for, what they need
• one for the operating system: the rituals, the cadence, how the machine actually runs
• one for the martech stack
• one for the strategic evidence – the research, the positioning work, what we tested and what it told us
• a vendor book
• and an index that maps every artifact, including the ones that live in SharePoint, in Slack, and in email

Then two front doors. A human one, a plain “start here” page. And a CLAUDE.md, which tells any future AI session how the brain is organized and how to keep it trustworthy: one source of truth per topic, absolute dates instead of relative ones, embargoes respected.

That second door is the part I’d argue for hardest. The person taking over doesn’t have to read the thing. They can ask it a question and get the real answer, with the reasoning attached. A new CEO can get an honest read on where marketing stands without me in the room.

Three things I’d take from it

Point AI at the judgment, not at the output. Everyone aims it at the deck, the doc, the farewell note. Polish is the cheapest thing these tools do. The expensive one is interrogation – sitting there while something patiently asks you why, until the reasoning you’ve been carrying around implicitly is on the page.

Structure beats prose. A memo ages badly, because it’s one person’s snapshot written in one mood on one Tuesday. A file that owns exactly one topic can be updated by someone who never met you. Six months from now, half of what I wrote will be wrong. The difference is that the wrong parts can be found and replaced instead of quietly rotting inside a fourteen-page narrative.

Building it made my own thinking sharper. You can’t hand over a decision you can’t articulate. There were a few I couldn’t – not because they were bad decisions, but because I’d never made the reasoning explicit, even to myself. That told me something.

I said to a colleague a few months ago that I’d gotten better at my job once I knew I was leaving. Some of that is just the fear being gone. But some of it is this: describing your own judgment to something that keeps asking why is a strange and useful mirror.

The document was never the point. What I wanted to leave behind was something the next person can argue with.

–S.

---

Company and colleague details have been anonymized.