Lab notebook
experiments with coding agents: hypothesis, instrument, number, and what went wrong
In reading order 1 post
- science technology engineering
I thought my harness was slow because it decided too much
I thought my agent harness was slow because the model made too many decisions. I measured a week of work with two instruments, and the time was somewhere else.
5 min read
Write to me
Something to add, dispute, or compare with your own work? Tell me by email, on LinkedIn or on WhatsApp.