Harnesses: Engineering AI Reading Group

​Theme: Harness Engineering in Coding and Beyond.

​Harness: The scaffolding around a model that controls how it receives tasks, uses tools, subagents, manages context, and checks its own outputs.

​As we observe in Terminal Bench, Harnesses can provide a lift of up to 3% on top of vanilla codex.

​Agents building Harnesses Autonomously:

• ​Meta-Harness: End-to-End Optimization of Model Harnesses

• ​Natural-Language Agent Harnesses

​Evals:

• ​Terminal Bench

• ​SWE-Bench

​Example Harnesses:

• ​Codex

• ​Pi

​Engineering AI RG comes together to read and code on tools, libraries and codebases that are used frequently in the modern AI stack.

​All are welcome.