Cognition Launches FrontierCode 2.0 to Test Coding Agents Beyond Detailed Specs

Cognition has released FrontierCode 2.0, a coding benchmark designed to assess whether AI agents can infer a user’s intent and produce changes that fit the wider needs of a codebase. The updated suite expands on the company’s earlier focus on whether maintainers would merge a proposed change, adding broader tasks, stricter grading and more emphasis on thoroughness. Cognition’s announcement describes the benchmark as an attempt to evaluate the user experience of delegating software work to AI agents.