Inside OpenAI’s Agentic Software Factory

X postGergely OroszNews

In an in-depth report, Gergely Orosz explores internal operations at OpenAI, where Codex and ChatGPT Work have evolved into the operational foundation across technical and non-technical teams. Powered by deep integrations with systems like Slack and Kubernetes, automated feedback loops like 'Perf Factory' are reshaping engineering workflows and organisational structures.

Key points

  1. Codex evolved within months from a nice-to-have utility into the operational backbone of OpenAI.
  2. Non-engineering departments (such as finance, legal, and recruitment) increased weekly Codex and ChatGPT Work usage from near 0% to approximately 90% without top-down mandates.
  3. The internal Codex harness connects to core infrastructure like Slack and Kubernetes, enabling agents to parse chat discussions or bug tickets and generate actionable implementations.
  4. Automated feedback loops, such as 'Perf Factory', continuously monitor production systems and trigger Codex agents to resolve performance bottlenecks autonomously.
  5. Traditional IDE usage has declined since January, prompting a re-evaluation of code reviews and pull requests as engineers rely on long-running multi-day '/goal' threads.
  6. Engineering specialisations are fading in favour of agency and high-level judgment, enabling individual engineers to complete large-scale migrations.

Why it matters

The report provides a rare, detailed look at how a frontier lab implements end-to-end agentic workflows in real production environments. Rather than using AI merely for in-IDE autocompletion, OpenAI demonstrates a model where integrated autonomous loops handle monitoring, remediation, and operational tasks across the organisation.

Analysis

Based on interviews with seven engineering leads at OpenAI, Orosz details how the integration of agent harnesses with enterprise infrastructure shifts engineering paradigms. The rapid internal adoption across non-technical staff suggests that sustained goal execution and system access drive utility more than polished interfaces. However, this model introduces new operational questions: pull requests, manual code reviews, and traditional IDE setups face structural disruption when multi-agent systems independently address long-running goals across distributed infrastructure.

Original post

Gergely Orosz

@GergelyOrosz · 15 September 2026

Here's what OpenAI's agentic software factory looks like, today. Details: t.co/UUZufLxG2r (thanks to all the OpenAI folks who explained how it works! And Perf Factory looks especially interesting to me) t.co/NEyldtr1Dq

4k likes129 replies6.5k bookmarks456.2k views

View on X
More posts in the thread (3)
  1. @spectragai Don’t just be focused on the prompt; OpenAI’s Codes is connected to every system, Slack etc Prompt could be whatever: “based on the convo in #channel make a plan how tim implement it, and show me” or “make a plan and build a version” or “based on the bug report ee discussed
  2. @jacobtechtavern And k8s and the most ironic one a11y
  3. @keeper87g9a5 AI slop Blocked
Selected replies (5)
  • @nwbotha @GergelyOrosz it's very impressive no doubt! But the diagram looks less sophisticated than what cursor/spacex is doing for instance with verification steps, etc. I assume a lot of the verification/complexity is in the rollout/monitoring step (with lots of details not in the diagram)
  • @timothyfcook @GergelyOrosz @nwbotha We're seeing the most advanced teams will have agents kick off automatic feature flags releases that monitor key guardrail metrics (via A/B test) during the progressive rollout. If it causes regression, hands off data+context to agent to autofix.
  • @DanHMcInerney @GergelyOrosz You can build this entire thing using only 2 core composable skills. t.co/qf44kDCWRx will combine those 2 skills into complex workflows. This is literally just a workflow. Plug and play upgrade to Codex or Claude Code.
  • @DanHMcInerney @GergelyOrosz I just went ahead and built this using t.co/qf44kDCWRx so you can "/software-factory build me a game that'll go viral on X" Took like, 3 minutes. Running A/B tests now to see how it performs. t.co/TnuECNHkzO
  • @realJacobJed @GergelyOrosz @grok is this real?

Mentioned links

Links and tools from this entry

Summary generated by AI (Gemini 3.8 Flash, 27 September 2026). It may contain errors – the original source is authoritative.

Related entries

Similar in content, found by the AI search.

  • Article:Gergely Orosz

    Inside OpenAI’s agentic software factory

    Gergely Orosz provides an inside look at OpenAI, where Codex and ChatGPT Work have transformed internal operations, shifting workflows away from IDEs to autonomous coding agents while a tenfold increase in PR volume forces a rethink of CI/CD.

    AI & LLMs· News

  • X post:Luna Tech

    OpenAI Workshop: Building Agentic Systems with Codex

    Luna Tech highlights a one-hour OpenAI workshop demonstrating how to build agentic systems with Codex from scratch, covering subagents, prompt patterns, persistent goals, and automated loops.

    4kbookmarks151.7kviews

    AI & LLMs· Resource

  • X post:Justin Duke

    Software Factories: Why the Value Lies in Org Process, Not Infrastructure

    Justin Duke highlights an essay by Cole Murray exploring the concept of the 'software factory'. Murray, maintainer of the open-source platform OpenInspect, argues that agentic development systems succeed or fail based on organisational processes rather than commoditised tooling.

    1kbookmarks124.8kviews

    AI & LLMs· Opinion

  • Video

    Video:CURT

    Harness Engineering: How to Build a Software Factory with AI Agents

    In his talk at the AI Engineer World's Fair, Tessl's Dru Knox introduces the concept of a software factory, where coding agents produce all user-facing code whilst engineers build and maintain the factory infrastructure. He outlines harness engineering (also referred to as loop engineering) as the guiding discipline, dividing engineering work into inner, outer, and meta loops to improve autonomy, automation, and overall code quality.

    AI & LLMs· Talk

Back to the knowledge base

Let's talk about your project

Get in touch

Locations

  • Mattersburg
    Johann Nepomuk Bergerstraße 7/2/14
    7210 Mattersburg, Austria
  • Vienna
    Ungargasse 64-66/3/404
    1030 Wien, Austria

Parts of this content were created with the assistance of AI.