Copasetic Flow

  This is kind of cool! I started a new project today using codex --yolo, (I didn't want to use my Claude tokens until I'd worked out the design and architecture a bit more.) When I got done coding with codex, I asked it to create a repo for the work we did. That seemed to work, but I wanted to verify that the repo was ok, so I typed  gt status and was immediately presented with  I answered y and the extension was off and running adding my new repo to Gas Town. I didn't actually have Gas Town up and running on my Linux client, so I had to get it started and then try again manually with  cd gt && gt rig add ssm_overlay https://github.com/hcarter333/ssm_overlay.git and it just worked!

 I've seen a few posts in the last week that talk about engineer's loosing their time to think because they're too busy watching what their LLM agents are doing. Anthropic's models, even under GasTown, have decide to spontaneously fork and find other ways to farm work out to subagents, occasionally causing agent storms on my project in the last week, so I can sympathize a bit more now that I did at the start of the week. There are a few solutions I'd like to propose for this issue. Here goes. How to Get Your Thinking Time Back First, if you're agents are doing marginally what you want, this is pretty simple. Stand up, back away from your desk, and leave . If it makes you feel better, setup a notification so the agent messages you when it's done. Perhaps, something like this If you don't trust your agents for the moment, still don't watch them. Come up with metrics to know if they worked or not and control your agents or modify their activities AUTOM...

 I'm still tracking down what makes some agents find a Nikola Tesla research finding and why others do not. Today, that's led me into investigating Claude CLI's forking harness. A few notes from me. Forking looks pretty spectacular! The agent kicks off a subagent that automatically has a copy of the parent's context. The subagent doesn't add to the parent's context until it's done. So, it seems to make things cheaper, at least for my passenger manifest research. The agent that used forking made the Tesla association. The other two agents with the same inputs and the same model did not use forking and did not find the Tesla assocation. This is important. It seems that agents that don't fork lack the persistence to look for more than one "really good" finding. They make that one good finding, and then kind of take any results for the rest of the passengs as good enough. Each forked subagent is looking for its own "really good" finding ...

 I'm working through a methodology to study the behavior of teams of agents via observation of real-world tasks. As usual with LLMs, the concept of repeatable results is squishy, especially as compared to non-LLM deterministic computing. My finding last week was that LLM agents, especially Claude (per Google's research), can exhibit stigmergic , (a fancy word for how insects, like ants, 'learn' where important locations are from other insects), learning and behavior. In short, agents given the exact same instructions, (prompts), can and often times will exihibit different behaviors if they can see the results of the work of other agents. If you want to study the variance in the behavior of an LLM agent over multiple runs, this stigmergic behavior has to be accounted for. Otherwise, we're not measuring the behavior of an LLM agent with a set of inputs and prompts. With stigmergic behavior, if we're not careful, we're observing the behavior of a community of ...

I finally got Beads working on Windows… sort of. The trick? I used WSL. I hadn’t tried to do serious work on Windows in quite a while, so I was genuinely surprised to discover that WSL (Windows Subsystem for Linux) just worked. I’d been struggling with Beads— a system that helps to automate workflows for LLM agents by essentially creating revision controlled context for them —for several days. Some of the things Beads wanted simply weren’t available on Windows. The big one was tmux (which I also needed for Gastown). I spent about three days fighting with that. Then I finally took the plunge and went all-in on WSL. And… it just worked. It worked especially well combined with OpenAI Codex. From WSL/Linux, I can fire up Codex CLI and tell it to install things for me. No worrying about which Windows installer to use. No chasing down extra packages. Everything just works. Need ffmpeg and Gimp ? You need merely ask.. I did have to start Codex CLI with: codex --yolo Then, with basic sudo ...