Presentation: Beyond the Harness: Canvas for AI Agents

| Source: InfoQ AI/ML

Tags: Nimisha Asthagiri, Thoughtworks, QCon AI, InfoQ, agentic software engineering, architecture drift, decision records

At QCon AI, Thoughtworks' Nimisha Asthagiri argued teams should move from rigid harness engineering to a bounded canvas that gives AI coding agents limited autonomy, drift detection, and decision records.

Details

InfoQ published Asthagiri's 49-minute talk, "Beyond the Harness: Canvas for AI Agents." Her argument applies established product engineering and architecture principles to agent-driven development. A bounded canvas grants agents autonomy within limits, adds continuous architecture drift detection, and keeps decision records so teams can manage cognitive debt as codebases grow. She draws on earlier experience leading architecture for an open-source community whose organically grown codebase suffered from stale feature toggles, unclear ownership, and a modular monolith sliding toward a big ball of mud.