Essays · Notes · Field reports
Long-form thinking on autonomous agents, the STORM standard, and what it takes to make agentic software that is composable, honest, and dull in all the right ways.
Why the next decade of agents will look less like magic and more like plumbing — and why that's the point.
A note on why STORM models autonomy as discrete cycles instead of agent chains, and what that buys you in production.
What we learned shipping STORM into a high-frequency execution loop — and the failure modes that only surface at scale.
Confidence scores, replay logs, and the case for agents that can be cross-examined.
Confidence-aware action commitment as a first-class primitive — not a wrapper, not an afterthought.
Why every STORM-conformant runtime ships the same four-step cycle — and what conformance actually means.
We pointed four specialist agents at the same hard question and watched them disagree. Here's what it taught us.
A list of features we cut, why we cut them, and what we'd add back only with overwhelming evidence.