ICAE-Bench evaluates coding agents as interactive project builders, starting with fuzzy product intent rather than a fully specified coding task.
Why it matters: For evaluating agents expected to build software from incomplete intent, the benchmark measures more than whether code passes functional tests: it also includes similarity, structure, design, and interaction quality.
Agent Framework Declarative Workflows 1.0 lets teams define multi-agent orchestration in YAML while loading it into the standard Workflow runtime.
Why it matters: Separating orchestration from application logic makes workflow behavior easier to review and lets teams diff, review, and ship many orchestration changes as YAML changes, while retaining the same Workflow runtime.