The record of what happened should compound up the ladder.

Stand-ups are performances.

The team recites status. The call ends. Nothing gets written down. Ask anyone an hour later what was decided - you'll get five different answers.

Nobody has a real picture of where things stand. What's blocked. What's actually moving. What was decided last week and why. The information exists somewhere, buried in Slack threads, scattered across Looms, living in the heads of whoever was in the room. But there's no single place it lives, and no way to get back to it.

That's not a communication problem. That's a capture problem.

So we built Pulse. It drafts your weekly update from the breadcrumbs the work already leaves - Slack, Looms, PRs, boards. You read it, edit a few lines, send.

Your manager gets a draft pulled from your update and everyone else's on the team. Their manager gets the same. Every level shapes what came before. The record builds up the ladder from the actual work - not from memory, not from what sounded good in a stand-up.

We used it with our own team. The draft was only as good as the signal that week - Slack activity, Looms, PRs. Some weeks plenty. Other weeks almost nothing. When the draft came back thin, it still felt like work. The routine never stuck.

We demoed to other teams. Everyone understood the value. Nobody wanted to be the person who had to get their whole team on board first. Not the writing - the coordination. That's what killed it.

It needed to work for one person before it could work for a team. One person starts. Gets value. The team grows into it.

AI's job isn't drafting. It's deciding.

What from this week's work belongs in the record. What gets skipped. How it connects to what came before. That's how the record compounds into something useful instead of just growing. Without that, you've got a pile of updates, not a history.

The model doesn't know when a decision is load-bearing until you tell it. That's still the human's call. But the record that makes that call possible - AI builds that.

I led product design. How updates become summaries, how context links back to source, how the rhythm stays light enough to hold. Ran demos. Worked close with engineering.

Six months. Then we stopped. The tool that was supposed to make the record automatic still required too much intention to keep alive. The capture had to disappear into the work. It didn't.