Presentation: Beyond the Harness: Canvas for AI Agents

InfoQ presents a talk titled "Beyond the Harness: Canvas for AI Agents." Nimisha Asthagiri, a Technical Advisor at Thoughtworks, shares insights on how applying product engineering and architecture principles can create a bounded canvas for AI agents. This model provides agents with bounded autonomy, continuous drift detection, and decision records to scale software development resiliently.

Nimisha discusses her personal experience with a similar challenge in an open-source community, where the codebase had grown organically, causing decay, feature toggles not being garbage collected, and code ownership confusion. Drift was also a significant issue as the monolith evolved into a big ball of mud. To support the growth of the codebase and community, guardrails, context, and guidance were required.

This experience is reminiscent of the current challenges faced with AI agents, especially as humans struggle to connect and innovate independently.

Nimisha introduces the concept of harness engineering, proposing a shift from a controlled harness to a bounded canvas. She emphasizes the importance of mental model shifts, accepting cognitive debt, and leaving traces or decision records to facilitate inheritance and work by others. Nimisha also highlights the relevance of chaos engineering in managing agentic topologies, even when not directly involved in the codebase.

Throughout the talk, Nimisha aims to share mental model shifts, propose a bounded canvas, and discuss evolutionary architecture, design for change, and deliberate disturbance of codebases and products. The talk concludes with a call for discussion and collaboration among attendees.