Deja Okonkwo

Agents need separate storage for ephemeral outputs while keeping project assets in Git LFS.

Automated policies replace human naming conventions to manage hundreds of agent-created branches.

Benchmarks designed for factory agents must measure process discipline, not just final correctness.

Trajectory-level evaluation catches agent failures that outcome metrics hide before merge.

Traditional velocity metrics hide AI-generated code decay behind inflated output numbers.

Organize system prompts in five layers to keep coding agents coherent across sessions.

Systems for automatically analyzing AI agent work logs at scale.

Transcripts reveal how coding agents actually behave and predict failures early.

Separating tool calling, tool search, and MCP prevents teams from optimizing the wrong layer.

Dead skills bloat AI agent prompts, wasting tokens and slowing inference with unused capabilities.