The meeting ends. The memory updates.

The worst thing about keeping a team memory is that someone has to keep it. The meeting ends, everyone goes back to work, and the decisions that were just made live only in a transcript nobody will reread.

This is exactly the job agents should have. Not answering questions — doing the maintenance.

How an agent works here

You pick an agent from the directory, point it at one of your memories, and give it what it needs — which source to watch, how often to run. From then on it runs by itself, on its schedule.

Take the meeting example. The agent reads the transcript and drafts what the memory should learn from it: the decision that was made, the deadline that moved, the new person who joined the project. Each of those becomes a proposed change.

Proposals, not silent writes

Here’s the part we care about most: the agent doesn’t get to write directly. Its drafts land in the same review queue as everything else, as proposed actions waiting for a person. You see what it wants to add, approve the good ones, drop the rest. An agent earns exactly as much trust as its proposals deserve.

The result is a memory that stays current without anyone having the “update the wiki” job. People make the calls. Agents do the typing.

The directory is open to browse — no account needed. If you’re wondering what teams automate first: meeting notes, then inboxes, then tickets. In that order, almost every time.