Google's Harness Engineering: Agent = Model + Harness Replaces Prompt Engineering
anselm · x · 2026-08-28
Google's team released a 9-page PDF introducing "Harness Engineering" as the skill replacing prompt engineering in 2026. The core formula is Agent = Model + Harness. The 6-step playbook includes:
- Add Guides: Convert past agent failures into permanent fix rules (AGENTS.md, rule files).
- Add Sensors: Implement linters, tests, and validation scripts for the agent to self-check output.
- Build the Agentic Loop: Plan, execute, verify, and fix with bounded retries and budget caps.
- Externalize Memory: The harness remembers state and artifacts across sessions where the model forgets.
- Enforce Permissions: The harness, not the model, enforces tool access, write limits, and approvals.
- Wire Observability (truncated in source).
More from coding & agent
- Archil Launches Persistent Sandboxes for Agentic Applications — ycombinator · 2026-08-28
- Grok Bot wired into Intercom and Cursor: like hiring another engineer — minchoi · 2026-08-28
- Grok Bot called a one-person company in your pocket: 10 real use cases — minchoi · 2026-08-28
- Speculative Programmatic Tool Calling: 2x Speedup for Agentic Code Execution — CShorten30 · 2026-08-28
- Are we paying a platform tax every time we build an AI agent? — rio_ARC · 2026-08-28
- Live Stream: Building Data Pipelines on the Fly — aronchick · 2026-08-28