Comment by sandinmyjoints
1 day ago
Lots of comments about Gas Town (which I get, it's hard not to talk about it!), but I thought this was a pretty good article -- nice job of summing up various questions and suggesting ways to think about them. I like this bit in particular:
> A more conservative, easier to consider, debate is: how close should the code be in agentic software development tools? How easy should it be to access? How often do we expect developers to edit it by hand?
> Framing this debate as an either/or – either you look at code or don’t, either you edit code by hand or you exclusively direct agents, either you’re the anti-AI-purist or the agentic-maxxer – is unhelpful.
> The right distance isn’t about what kind of person you are or what you believe about AI capabilities in the current moment. How far away you step from the syntax shifts based on what you’re building, who you’re building with, and what happens when things go wrong.
> either you look at code or don’t, either you edit code by hand or you exclusively direct agents, either you’re the anti-AI-purist or the agentic-maxxer – is unhelpful.
If you're looking at all your code you are just walking the motorcycle. You need tests to automate your eyes. In fact I believe tests and specs are the new product, code can be regenerated at will.
That is why we see vibe coding projects that replicate well specced and implemented products like web browsers, you get both the specs and differential testing for free.
> Buried in the chaos are sketches of future agent orchestration patterns
I'm not sure if there are that many. We need to be vigilant of "it feels useful & powerful", because it's so easy to feel that way.
When I write complex plans, I can tell Claude to spawn agents for each task and I can successfully 1-shot a 30-60 minute implementation.
I've toyed with more complicated patterns, but unlike this speculative fiction, I did need my result both simple and working.
A couple of times now I've had to spend a lot of hours trying to unfuck a design i let slip through. The kind where 1 agent injects some duplicate code/architecture pattern into the system that's correct enough not to be flagged, but wrong enough to forever trip up every subsequent fresh agents that stumble on it.
I tell people my job now is to kick these things every 15 minutes. Its a kinda joke kinda not. But they definitely need kicking. Without, the decoherence of a non-trivial project is too high, and you still need time to know; where and how to kick.
I'm not sure what I'd need to be convinced a higher level of orchestration can do that. I do like to try new things. But my spider-sense is telling me this is a Collatz-conjecture-esque dead-end. People get the feeling of making giant leaps of progress, which anybody using these things should be familiar with by now, but something valuable is always just out of reach with the tools we currently have.
There are some big gains by guiding agents/users to use more sub agents with a clean context - perhaps with some more knobs - but I'd advise against acting under the assumption using grander orchestration tools will inevitably have a positive ROI.