← @bgSponge

Inside OpenAI’s agentic software factory

Summary

Gergely Orosz shows how OpenAI uses Codex to build an agentic software factory, replacing traditional IDEs and manual pull requests with automated feedback loops and domain-specialist agents.

  1. Codex replaces traditional developer tools.

    IDE usage has declined since January as Codex became the backbone for all workflows, including non-engineering tasks like finance and legal.

    everything is now a coding agent

  2. Development acceleration breaks existing infrastructure.

    Pull requests per engineer are growing like a hockey stick, causing a 10x load increase on build-test-deploy pipelines within six months.

    Every month, we wake up to a new set of infrastructure scaling challenges

  3. Agents reimagine the code review process.

    OpenAI uses multiple domain-specialist agents to review changes, simulating the presence of infrastructure and security experts for every pull request.

  4. The software factory automates deployment monitoring.

    Agents handhold changes to production by building their own monitoring dashboards and watching relevant signals to ensure safe rollouts.

    Builds its own monitoring dashboard to use

  5. Perf Factory closes the feedback loop.

    The system uses agents to sift through alerts, identify latency regressions, and propose fixes to maintain continuous improvement.

    identify real latency regressions, root-cause them and propose fixes.

Highlights

  • The big theme of the past months has been that everything is now a coding agent.

  • We’re talking about roughly a 10x increase in load on some systems.

  • Codex being good at longer-running tasks seems to cause people to do fewer things in parallel.

  • the way we do code review today makes less and less sense, and the same is true for pull requests.

  • The big theme of the past months has been that everything is now a coding agent. Whether the visible code is your output or not

  • The agent builds the code, runs the tests, fixes the code when it breaks the test, and then creates a pull request.

Read the original ↗Save to your Sponge

Summarized with Sponge. Quotes are the author’s own words.

More from @bg

All readings →