AIAug 29, 2026
Long-running agents need durable state, not immortal containers
Reliable agents recover because their tasks, evidence, and permissions outlive the worker that happens to be executing them.
Read moreWriting
Research notes, technical writing, and industry perspectives on AI, control, and energy systems.
AIAug 28, 2026
Harness Engineering: The System Around the Model
Agent performance is not only a model problem. Harness engineering designs the context, tools, controls, verification, and learning loops that make an agent dependable in the real world.
Read more
EngineeringMay 23, 2026
Image-to-Code: Designing Frontends Before You Write Them
AI-generated websites converge on the same gradient-blob template because we're asking coding agents to invent visual taste while they're also handling implementation. A better workflow: generate the design first, then code section by section against the reference. A synthesis of Leon Lin's image-to-code thread, with notes from doing this on the portfolio you're reading now.
Read more