📢 Time for an update on my workflow. This one's a 23 min read, so buckle up. 100% organic and hand-written, like an animal. steipete.me/posts/just-talk-…
Jupp.
Interestingly enough, the workflow you describe is 98% the workflow I have with Claude.
I don't get how we both can have the opposite experience.
This is good stuff, and I appreciate the feedback about Amp! If a single model house has all the best coding models, then yes, they will win. But if it ends up being valuable to use a variety of models, then things like Amp can win. We'll see.
Incredible, very detailed and opinionated, thanks!
I think the main things I probably think of a bit more are using the testing out model SDKs as part of the workflow, trying to design agentic workflows and feedback loops with other models.
>IMO most are something for the marketing department to make a checkbox and be proud. Almost all MCPs really should be clis.
Lots of great knowledge in this post (and things that needed to be said out loud), thanks for writing this
so @kbal11 runs the @latentspacepod AI in Action discord club (every friday 1pm) and we love this kind of convo
wondering if you'd be up for joining to talk long form thru this?
Very cool- interesting point about clis - for my tool (ck (“seek”) a sidecar local embedding hybrid vector bm25 code search) - I found it was pretty good just calling it directly rather than using the MCP version github.com/BeaconBay/ck
This is really good stuff Peter. Couldn't agree more with the "just talk to it" takeway. I cringe at the amount of context poisoning I see some people suggest.
Just like the old "I do multi-tasking well" approach caused havoc to mental health (because human brains dont do multi tasking well), I feel like this idea of running 3-8 agents parallelly doing different tasks, can really stress the developer. Do you not find it stressful?