Chaitanya Meesalahttps://fermibubble.github.io/Writing and notes on autonomous systems, agent infrastructure, rollout intelligence, and reliable software.Trustworthy Autonomyhttps://fermibubble.github.io/writing/trustworthy-autonomy/https://fermibubble.github.io/writing/trustworthy-autonomy/Thu, 03 Sep 2026 12:00:00 GMTTrust is not a personality trait of the model. It is a property of the system surrounding it.What does “good” mean for an agent?https://fermibubble.github.io/notes/what-does-good-mean-for-an-agent/https://fermibubble.github.io/notes/what-does-good-mean-for-an-agent/Thu, 03 Sep 2026 12:00:00 GMTAI exposes how much of quality was previously implicit infrastructure.The Agent Harness Is the Producthttps://fermibubble.github.io/writing/agent-harness-is-the-product/https://fermibubble.github.io/writing/agent-harness-is-the-product/Sat, 29 Aug 2026 12:00:00 GMTThe model supplies general intelligence. The harness turns that intelligence into dependable work in a particular world.Skills aren’t policieshttps://fermibubble.github.io/notes/skills-arent-policies/https://fermibubble.github.io/notes/skills-arent-policies/Fri, 28 Aug 2026 12:00:00 GMTProcedure, authority, and enforcement are different concerns.Memory should accumulate judgmenthttps://fermibubble.github.io/notes/memory-should-accumulate-judgment/https://fermibubble.github.io/notes/memory-should-accumulate-judgment/Fri, 21 Aug 2026 12:00:00 GMTTranscripts preserve history. Useful memory changes the next decision.Rollouts Are Decision Problemshttps://fermibubble.github.io/writing/rollouts-are-decision-problems/https://fermibubble.github.io/writing/rollouts-are-decision-problems/Tue, 18 Aug 2026 12:00:00 GMTA rollout reviewer should not merely detect anomalies. It should decide what the available evidence justifies doing next.Context engineering vs. harness engineeringhttps://fermibubble.github.io/notes/context-vs-harness-engineering/https://fermibubble.github.io/notes/context-vs-harness-engineering/Fri, 14 Aug 2026 12:00:00 GMTContext makes the world legible; the harness makes action dependable.Why agent verifiers failhttps://fermibubble.github.io/notes/why-agent-verifiers-fail/https://fermibubble.github.io/notes/why-agent-verifiers-fail/Fri, 07 Aug 2026 12:00:00 GMTA second model does not automatically create independent assurance.