The decision got made in a thread. Then it vanished.

Two people argue for nine messages in a chat thread. Someone proposes a number. Someone pushes back. A third person says “fine, let’s do that” with a thumbs-up. The decision is made. Real money or a real promise now hangs on it. Then the thread scrolls, the channel moves on, and a week later the spec still says the old number, because nobody went back to change it.

Now the company has two answers to the same question. The doc says one thing. The thread, if you can find it, says another. And the person who has to act on the decision picks whichever one they happened to see.

A decision gets made where the conversation happens, and recorded, if at all, somewhere else. That gap is where companies forget what they already decided.

The decision and the record live in two different places

Here is the thing nobody designed on purpose but every company runs anyway.

Decisions are made in conversation. They’re made in a chat thread at 4pm, in the last five minutes of a call, in a reply to a reply that three people saw. That’s not a failure of discipline, it’s just where humans decide things. The arguing, the trade-off, the moment someone says “okay, do it”, all of that is verbal and fast and lives in a stream of messages.

Records live somewhere else entirely. The spec, the doc, the project board, the wiki page. Slow, structured, deliberate. Built to be read later by someone who wasn’t there.

The two are connected by exactly one thing: a human remembering to walk the decision from the place it was made to the place it’s supposed to be kept. And that walk is the most-skipped step in any company, because at the moment the decision is made, it doesn’t feel undone. It feels finished. Everyone in the thread agrees. Why would you stop the momentum to go update a document?

So you don’t. And the record quietly goes stale the instant the decision is made.

The naive fix: write better docs. It fails the same way.

The obvious answer is discipline. Write a decision log. Make people update the doc. Add a step to the process: every decision gets recorded.

It fails for a reason that has nothing to do with how good your people are. The recording step happens after the decision and somewhere other than where the decision was made. You’re asking a tired human, in the wrong tool, to re-derive what was just agreed and transcribe it into a structured place. That’s a second job, done at the worst moment, with no immediate reward. The decision already feels done. The transcription is pure tax.

And even when someone does it, the link is gone. The doc now says “we decided X.” It doesn’t say who decided, when, why, or what we rejected. Six weeks later someone reads “we decided X,” disagrees, and has no way to know whether X was a careful trade-off or a typo. So they relitigate it. The decision gets made a second time, possibly differently, and now you have three records that disagree.

The naive fix treats forgetting as a willpower problem. It isn’t. It’s a location problem. The record is in the wrong place to ever be trustworthy.

A decision recorded somewhere other than where it was made is a decision waiting to be forgotten.

The fix is not to try harder at the walk. It’s to delete the walk.

Capture the decision where it’s made, not where it’s filed

The key idea is simple. If decisions are made in conversation, then conversation is where they have to be captured. Not transcribed later. Captured there, as it happens, by something that was in the room the whole time.

It’s worth explaining why that changes the problem completely.

A company brain, a system that reads the conversation as it flows, doesn’t need a human to walk anything anywhere, because it never left the room. When the thread reaches “okay, let’s do that,” the brain already has the nine messages of context that produced it. It saw the proposed number, the pushback, the trade-off, the moment of agreement. It can capture all of it as one structured fact: this was decided, on this date, in this thread, with this reason, and here is the option we rejected.

The decision and its record become the same event instead of two events with a fragile human bridge between them.

Now compare the failure modes. The naive doc says “we decided X” with no provenance, so it gets relitigated. The captured decision says “we decided X on Tuesday, because of Y, having considered and rejected Z, here’s the thread.” When someone wants to reopen it, they’re not arguing in the dark. They’re arguing with the actual reasoning, in front of them, with the rejected alternative already named. Half the relitigation never happens, because the answer to “wait, why did we do this?” is already attached to the decision.

That’s the whole move. Stop asking humans to copy decisions into a system of record. Make the system of record present at the moment of the decision.

The contradiction is the alarm, not the embarrassment

There’s a second failure the naive approach can’t even see: the contradiction itself.

When the thread says one thing and the doc says another, nobody notices until it bites. Someone acts on the doc. Someone else acts on the thread. The two collide downstream, a wrong number ships, a promise is made that the team already walked back, two people show up with two different plans. The contradiction existed for weeks. It just lived nowhere that could see it.

The naive fix can’t catch this because the two records don’t know about each other. The doc has no idea a thread disagrees with it. The thread scrolled away and forgot it ever said anything.

A brain that captured both can. The moment a new decision in a thread contradicts a recorded one, that’s not an awkward inconsistency to be quietly buried, it’s a signal worth surfacing. Heads up: this thread just decided something that contradicts what the spec says. One of these is stale. Which one wins? The contradiction becomes a question you get to answer on purpose, on Tuesday, instead of a collision you discover on Friday when it’s already shipped.

This is the part people underrate. The value isn’t only that decisions get recorded. It’s that disagreements between records get caught, because for the first time, the records live in one place that can compare them.

What you get when decisions stop evaporating

Put the pieces together and the shape is clear.

Decisions are captured where they’re made, not walked by hand to a doc that goes stale. Each one carries its reason and its rejected alternative, so it survives the question “why did we do this?” without being relitigated from scratch. And when two records disagree, the disagreement surfaces as an early question instead of a late collision.

None of this requires anyone to be more disciplined. It requires the system of record to stop being a separate place you have to remember to visit, and start being present in the conversation where the deciding actually happens. The discipline moves out of the human and into the structure, which is the only place discipline ever survives a busy week.

The turn: a company that can’t forget what it decided

Here’s the part that isn’t about software.

The cost of a forgotten decision is rarely the decision itself. It’s the second meeting to make it again. It’s the new hire who reads the stale doc and builds the wrong thing in good faith. It’s the founder who has to be the human index of every choice the company ever made, because the choices live in a hundred threads only they remember being in. A company’s memory of its own decisions ends up stored in the few people who were present for them, and the day one of those people is busy, or on vacation, or gone, that memory goes with them.

That’s the real tax. Not the minutes spent updating a doc. The slow erosion where a company keeps re-deciding things it already settled, because it has no reliable way to know what it settled or why. Every re-decision is hours that could have gone to the decisions you haven’t made yet, the ones only a human should make.

A company that remembers what it decided, and why, and what it turned down, gets to spend its judgment on new questions instead of old ones. That’s the difference between a team that compounds and a team that keeps paying the same tax on Tuesday after Tuesday.

---

This is what we’re building at Apollo, not a better place to file decisions after the fact, but a company brain that’s present where the deciding happens, so the record and the decision are one thing instead of two. If you’ve ever found a thread that contradicts your own spec and couldn’t remember which one was right, you already know which one your company has been trusting: whichever one you happened to see first.