This last 30 days I have 3 agents running full time in the background doing specific tasks for me. Openclaw is the generalist one, it runs all experiments, all dirty things I just prompt and fuck it. It is my entry point. When I am golden about what I want I code a structured one
Testing cmux today. I'm pretty impressed at the moment. I've been, for a long time looking for a way to keep all my terminals in the same window so I can make this mix fine with aerospace.
Some small features that make is worth it: 🧵
One side effect of being an AI first company is that everyone will start pinging your API through claude or whatever harness you want in the most crazy ways.
Even if the API is internal only.
Having safety nets is more important than ever.
Agents have made documentation more valuable than ever. We document not only how the application is supposed to behave but how the new code is supposed to be written. That is more important than ever.
I have the feeling that the more I work with agents the more my job is to find a good feeback loop. A feedback loop is just to define the initial state, define how to iterate to the goal and define how to find if you are closer to that goal or further.
https://t.co/sMhv9X9rud