The most eager people I knew were leaning into Aider this time last year. It felt premature to me - jagged, tricky to use, more friction than flow.

Now everyone’s on Claude Code.

Paul Gauthier, Aider’s creator, was early. He set things up for everyone else. That’s how these things go.

The next question

Running multiple agents at once - orchestrating them rather than managing one at a time - is becoming a topic lately. Chris Laidler, one of the best people I know at staying current with this stuff, has been thinking about it: “I don’t believe agents running agents is ready yet.”

Steve Yegge just released Gas Town, an orchestrator for Claude Code. Even he includes this warning:

Do not use Gas Town if you do not juggle at least five Claude Codes at once, daily. Do not use Gas Town if you care about money. Do not use Gas Town if you are more than 4 feet tall. I want to tower impressively at meet-ups, like Sauron. Do not use Gas Town.

(This is almost the most Yegge quote imaginable.)

If Chris thinks it’s not ready, that’s information about how far we have to go. The big wrapper companies - Cursor, Windsurf, the rest - have almost certainly been experimenting with orchestration longer than Yegge’s pet project. Nothing shipped yet. That silence says something.

Chris Dixon’s line comes to mind: “What the smartest people do on the weekends is what everyone else will do during the week in ten years.” Gas Town isn’t literally a weekend project - Yegge left Sourcegraph and is working on it full-time - but it’s in that spirit. And the people experimenting with it are very much in weekend project territory. That’s worth paying attention to.

The feeling

This feels like March 2024 for agents - but one rung up. The timelines are accelerating, so I don’t have a strong feeling about when the tipping point comes. I can just sense that smart people are starting to ask the natural next questions.

The abstraction ladder keeps extending. Each rung seems less than serious until suddenly it’s ubiquitous.

The timing problem

There’s a more general lesson here about managing exploratory tool usage. Wait until things are mature and you’ll be late. But you can also burn a lot of time and energy checking whether things are almost ready.

I don’t have a good answer to this. But if you’re a skunk works team within your organisation - or if you’re trying to be the inkblot yourself - it makes sense to be continually asking these questions. Somebody has to stay close enough to know when the tipping point arrives.