Why I built Laterbase

Something changed when I started doing most of my coding with agents. They read a lot more code than I ever did. Ask one to fix a checkout bug and it'll wander through the payment client, the webhook handler and the retry logic on the way, and it notices things. A retry that swallows errors. A test that never actually runs. A config value that's been wrong for months.

None of it has anything to do with the task. So it ends up as a line at the bottom of the agent's reply: "By the way, I noticed…". And then we both move on.

Where it all went

For a while I tried to keep up with it. It didn't go well.

Chat threads. Half my findings lived in old sessions. I'd remember there was something about the webhook retries, but not which thread it was in, and by the time I found it the context was gone.

TODOs. I'd write a TODO, in the code or in a notes file, and that was the last anyone saw of it. A TODO is a promise to remember, and I don't.

My head. The worst one. "I'll come back to that" works fine for about a day.

And the bit that really got me: every new session started from zero. I'd spin up an agent to work on payments and watch it rediscover the exact problems another session had already mapped out the week before: same files, same digging, same conclusion. All that work, done twice, because the first copy was sitting in a thread I'd closed.

What I actually wanted

I didn't want another issue tracker. I've never once opened Jira in the middle of a session to write up something an agent spotted, and I'm not going to start.

What I wanted was simple:

• When something gets left for later, it gets written down. Straight away. Whether the agent spots it or I'm the one saying "refunds can wait, just get checkout working".
• The agent does the writing. It already knows the files, what it was doing and what it found. Why would I retype that?
• It comes back when it matters. Next time anyone starts work on that code, they hear about it first, before they redo the digging.

That's Laterbase. When an agent finds something out of scope, or I put something off, it pins it without asking and mentions it in a line. Each pin carries what's wrong, where, how it came up, the traps waiting for whoever fixes it, and how it would go about the fix. When a new session starts on the same files, it checks what's already known there first. And when I've got a spare hour and ask what's worth doing, I get a short list instead of a scroll through old threads.

How it's going

I've been dogfooding it while building it, which has been a bit circular in a fun way: a lot of what's in the backlog right now was found by agents working on the backlog.

The difference is mostly that I've stopped worrying about losing things. The agents find stuff, it gets pinned, and I deal with it when I'm ready. It's still early, and I'm sure there's plenty my own agents haven't found yet. Pinned, presumably.