Blog
Long-form write-ups on building AI systems — agent design, self-improving loops, and the infrastructure decisions behind these projects.
July 6, 2026
Ratchet Loops: Making an AI Agent Improve Your System Without Letting It Grade Its Own HomeworkA builder's guide to ratchet self-improving loops: agent sessions where every improvement is backed by a before/after score, code rolls back, the world gets compensated, and the loop can never lower its own standards. Includes two downloadable Claude skills.