Make your AI agent state its assumptions
AI agents invent plausible premises and write confident code on top of them. Here is the config rule that forces the assumptions into the file first.
— NANOMAGNETISM · OPEN SCIENCE · DEV WORK
Slow physics,
written down.
Computational physics PhD. I simulate magnetic nanoparticles at scale, and write about the workflows around the work: simulations, Linux, scientific computing.
— WHAT I WORK ON
Three things I spend too much time thinking about.
- 01
Nanomagnetism
Exploring how magnetic materials behave at the nanoscale, where disorder, noise, and frustration produce surprisingly rich physics.
- 02
Spin Glass Physics
Simulating assemblies of magnetic nanoparticles that freeze into disordered states. What looks like chaos turns out to be deeply structured.
- 03
Workflows & Tools
Simulation code, GPU kernels, and the workflows around them. Less typing every line now, more directing and correcting. The output is what counts.
-
-
Make your AI agent's rules executable, not just documented
A documented rule your AI coding agent keeps overlooking is a rule you should make executable. Here is why a lint-style hook on every edit fixes it for good.
-
Why AI sent me back to paper
The more work I hand to AI agents, the more notepads I use up. On the bottleneck moving from doing to deciding, and how paper became the overflow buffer.