Your old AI agent conversations areworth more to the next agent than to you

Most people never open an AI conversation again once it's over. The transcript is still useful, and the one who gets the most out of it is the next agent, not you.

Three agents, killed mid-file

Last Thursday I told an agent to stop at a good stopping point, because I was about to hit my usage limit. It had three other agents writing code for it, and it stopped them the fast way.

It killed all three mid-file.

One of the last notifications read "Now the job detail page and its components." Whatever they knew about what was done and what was left looked gone.

It wasn't. A killed agent keeps its transcript, and Claude Code can resume a subagent when you send it a message. So I sent each one a message: build nothing else, write down what you finished, what's half-done and what's next, then commit it.

All three came back. The notes ran 61 to 150 lines, and they held things no file on disk showed: an auth conflict with one video API, a quirk in how another video API takes its input, and a $10 spending cap that would have rejected every test job I'd set up.

What I should have asked for was a wind-down, not a kill. Start nothing new, finish the current file, commit with notes on what's done and what's next. A wind-down costs a few minutes of agent time. A kill costs the next session working out again everything the agents knew.

Anthropic's own write-up on harnesses for long-running agents points the same way: have the agent keep a progress file and commit as it goes, so a stop is a formality.

The old sessions aren't for me anyway

Most people never open a conversation again once it's over. Neither do I, mostly.

I've come around to thinking the old sessions aren't for me anyway. They're for the next agent, which will happily read through my last 30 days without getting bored. As of this morning that's 5,740 transcripts on this machine.

Here's how else I use them now.

Six ways I put old sessions to work

1. Reuse a fix. When a new agent hits a problem an earlier one already solved, I point it at that session and tell it to fix it the same way.

2. Ask my second brain. My sessions get copied into my second brain, a database I can query. There are 1,892 of them in it since the end of June, so I can ask what I worked on on a given day and what I learned.

3. Turn friction into skills. I have agents read a week of sessions looking for the places I got stuck or had to correct them twice, and turn those into skills (saved instructions my agents load) and memory. The next session starts a little smarter.

A week is more than it sounds. One sweep I ran in August read 339 sessions across seven days and ran 149 agents to do it.

4. Find it by describing it. When I can't find a conversation, I describe what we did and the agent finds it. I never remember titles.

5. Let sessions talk. An agent in one session can message an agent in another to work on something together. It doesn't always land.

Two weeks ago, two messages I thought were delivered sat in the other session waiting for me to click approve, and expired before I looked. The tool had reported both as sent. Now a message doesn't count as sent until the other agent confirms it got it.

6. Audit what gets used. Earlier this month I counted which skills my agents actually call, straight from the transcripts. They had called 101 different skills in a month. Of the 163 I had kept, 75 hadn't been called once in 30 days.

A transcript is a record of what was said at one moment

That's the one rule I had to write down before any of this worked. A transcript is not evidence of where anything stands now. "I will finish this tomorrow" is a lead, not a fact.

So when an agent reads old sessions to work out what's unfinished, it checks the actual artifact before it calls anything unfinished. The commit. The branch. The file. The live URL.

The killed agents are the same rule from the other side. Their transcripts knew things the files didn't, and the fix was to get those things written into a commit, where the next agent can check them.

Anyway.

I still rarely reopen an old conversation myself. My agents do it all day, and last Thursday three of them reopened their own.

What's something you worked out with an AI that you'd hate to have to figure out again from scratch?