How to build a shared universal AI harness?
Every project that uses LLMs for generating code, tests, or documentation should include some kind of harness – a set of rules that defines how the agent should behave. Unfortunately, the configuration of these rules is still inconsistent; the individual cloud AI tools (Copilot, Codex, Claude) each use different formats. Our team is made up of people from different organizations, so we set out to create a harness that would be common to all tools and lead to similar agent behavior.
