Matthew Berman requests software feature-building loops, prompting Peter Steinberger to note they require clearly specified target issues

@steipete Have you built features with loops?

@MatthewBerman Summary if you want to make a video 🙌

@MatthewBerman mine, from today, fully public: spec → plan → fresh subagent per task on cheap models → two adversarial reviews per task on the big one → merge at green.

9 hours, 30 subagents, 13 PRs merged.

the loop shipped the fix for its own routing: http://ax.necmttn.com/s/Necmttn/36d340286c8d43d1d68d9e5e0e36f2fc

@MatthewBerman if the issue is clear and specified, yah

@MatthewBerman /goal Please implement PROBLEM_STATEMENT. See linux, xnu, and llvm-project directories to ground yourself. I want performant and easy to read LLVM IR complete with unit/integration tests and profiling benchmarks. Pay special attention to cache use and IO overhead.

@MatthewBerman

@MatthewBerman I missed some updates Matt #spacex just sold my house, my wife doesn’t know

@steipete Can you provide an example? I’m trying to put together some educational content on loops

@RXed_EU @MatthewBerman Just tell her......

@goldenelephant1 @MatthewBerman She’s a lawyer, lawyers don’t believe in aliens @CarloDAngelo right?

@MatthewBerman here is how it looks

@MatthewBerman I would only use "loops" on a self hosted/local AI model.

@MatthewBerman is it just me, or are loops overkill for day-to-day dev? I find myself quite happy with a strong planning skill and execution skill.

@MatthewBerman

@MatthewBerman We are building them now for lemmebuyit and skill sets on the MCP server.

@MatthewBerman sorry, no. proprietary knowledge. 😆

@MatthewBerman 10 PRINT "CREATE SLOP" 20 PRINT "BURN TOKENS" 30 GOTO 10 RUN

@MatthewBerman Squirrel!

@MatthewBerman Have quite a few… the one I like best is how to set up tools in @LangChain has reduced cycle time dramatically