Warp builds self-improving agents on Claude

Warp, an AI-powered terminal platform, developed a self-improving agent framework using Claude that converts user feedback into continuous refinement loops through a "skills-based" architecture. The system stores knowledge in file-based skills rather than raw prompts, allowing agents to learn from human feedback between sessions and compound improvements over time, solving the problem of low-quality agent outputs that plagued their initial code review agent. Warp has integrated this approach across its platform, which now runs over 400,000 Claude Code sessions weekly and serves 800,000 developers.