Blog
Long-form technical writing on engineering, design, and the AI work shipping behind this site.
Loop Engineering vs. Harness Engineering: The Two Most Confused Concepts in AI DevelopmentEN中文
Harness Engineering and Loop Engineering get conflated, but they solve different problems: a Harness is the working environment — terminal, Git, browser, screenshots, Figma, Playwright, design systems, MCP servers — that lets an agent see, touch, and operate, while a Loop is the self-correcting PLAN → EXECUTE → VERIFY → REPLAN workflow built from a goal, state, verification, feedback, and stop conditions. They're complementary, not competing: a Harness solves discovery, a Loop solves optimization. Loops only pay off once an objective verifier exists, so in ambiguous, design-less early work — where requirements keep shifting — a Harness plus human-in-the-loop matters far more than racing ahead in a Loop.
Loop Engineering in Depth: State, Verification, and Autonomous AgentsEN中文
A deep dive into loop engineering — why state, verification, and feedback matter more than the model, when an autonomous loop is worth building, and the right sequence for taking one from a manual run to full automation.