Harness Engineering: Leveraging Codex in an Agent-First World

OpenAI describes a five-month experiment where a team built and shipped an internal software product with zero manually-written code — every line was generated by Codex agents, resulting in roughly a million lines of code and 1,500 merged PRs. The engineering role shifted from writing code to designing environments, specifying intent, and building feedback loops: making the application legible to agents via Chrome DevTools integration, a full local observability stack (logs, metrics, traces), and a structured in-repo knowledge base that replaces monolithic AGENTS.md files. Key lessons include enforcing architecture through custom linters and structural tests rather than micromanaging implementations, treating repository-local versioned artifacts as the single source of truth (anything not in the repo “doesn’t exist” for agents), and running automated “garbage collection” processes to continuously pay down technical debt. The team achieved an average of 3.5 PRs per engineer per day and estimates they built the product in about 1/10th the time manual coding would have required.

Source: Harness engineering: leveraging Codex in an agent-first world