Source: https://internal-agents.com/agents/openai-software-factory # OpenAI — Agentic software factory An agent system built around OpenAI's internal Codex. A person defines the outcome; Codex implements the change and babysits CI until green; domain-specialist agents review each change behind risk-tiered routing; and a per-change deploy agent handholds approved changes to production and builds its own monitoring dashboards. - Approach type: Agent system - Deployment stage: Scaled - Autonomy: Human in loop - Evidence strength: Mixed - Status: Internal - First reported year: 2026 - Work: Coding, Code review, CI triage, Ops - Interfaces: Desktop, Cli, Slack, Github, Skill - Invocation: Interactive, Background, Scheduled, Event driven - Entry reviewed: 2026-09-16 ## Where people stay involved Each scope pairs its normal attention boundary with supporting evidence. See the [supervision definitions](https://internal-agents.com/definitions#supervision) for the level mapping and limits. - **low-risk pull request -> merge in opted-in codebase areas** — Exception only · Level 5 - **code change -> production rollout on the general path** — Work product review · Level 3 - **production alert -> proposed performance fix** — Unknown · Level unknown ## Overview ### Summary An agent system built around OpenAI's internal Codex. A person defines the outcome; Codex implements the change and babysits CI until green; domain-specialist agents review each change behind risk-tiered routing; and a per-change deploy agent handholds approved changes to production and builds its own monitoring dashboards. Fact · Reported · Medium confidence · `openai-software-factory--summary` Confidence reason: A linked participant or independent source reports the claim. Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, lines 138-186 (pipeline steps 1-8) - Contextualizes · [3] [Harness engineering: leveraging Codex in an agent-first world](https://openai.com/index/harness-engineering/) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-harness-engineering-post/content.md) · Preserved content.md, lines 46-48 (one OpenAI team's repository runs review predominantly agent-to-agent; a single team's account, not the company-wide pipeline) ## How it works ### Supporting component Multiple agents, each configured as a domain specialist, review every change; the article compares this to a review by a domain expert from each relevant infrastructure team Fact · Reported · High confidence · `openai-software-factory--primitives-0` Confidence reason: A linked first-party source states the claim. Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, lines 155-157 (domain-specialist review agents) - Supports · [2] [The most important OpenAI announcement you probably missed at DevDay 2025](https://venturebeat.com/infrastructure/the-most-important-openai-announcement-you-probably-missed-at-devday-2025) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/venturebeat-devday-codex/content.md) · Preserved content.md, line 44 (DevDay 2025, October 2025: nearly every pull request at OpenAI is reviewed by Codex, catching hundreds of issues daily; OpenAI's own figure relayed by press) - Supports · [3] [Harness engineering: leveraging Codex in an agent-first world](https://openai.com/index/harness-engineering/) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-harness-engineering-post/content.md) · Preserved content.md, lines 46 and 48 (Codex requests additional specific agent reviews locally and in the cloud until all agent reviewers are satisfied; 'almost all review effort' handled agent-to-agent; one repository) ### Supporting component Changes are classified by risk; high-risk changes can trigger more agent reviews or a mandated human review, while opted-in low-risk areas use an agent that auto-approves pull requests Fact · Reported · Medium confidence · `openai-software-factory--primitives-1` Confidence reason: A linked participant or independent source reports the claim. Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, line 159 (risk classification and auto-approve) ### Supporting component After a human approves production, an assigned agent handholds the change to full rollout; it decides which signals mean success or failure, builds its own monitoring dashboard, and watches production signals Fact · Reported · Medium confidence · `openai-software-factory--primitives-2` Confidence reason: A linked participant or independent source reports the claim. Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, lines 165-175 (agentic deploy) - Contextualizes · [6] [zbrock: many internal teams adopted the same practices](https://news.ycombinator.com/item?id=48435213) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/hn-zbrock-internal-adoption/content.md) · Preserved content.md, lines 12 and 16 (harness-engineering co-author: many internal teams adopted the same practices; some run centralized agent-mediated integration queues and local Codex threads that monitor CI; no self-built deploy dashboards described) ### Supporting component A perf harness sends problematic pull requests to the Synthetics A/B framework to evaluate performance implications Fact · Reported · Medium confidence · `openai-software-factory--primitives-3` Confidence reason: A linked participant or independent source reports the claim. Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, line 153 (perf harness sends problematic PRs to the Synthetics A/B framework) ### Supporting component Agents sift through alerts and dashboards, de-duplicate signals, identify real latency regressions, root-cause them, and propose fixes Fact · Reported · Medium confidence · `openai-software-factory--primitives-4` Confidence reason: A linked participant or independent source reports the claim. Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, line 186 (Perf Factory) ## Supervision evidence ### Operating model assessment Level 5 for low-risk pull request -> merge in opted-in codebase areas; human attention boundary: exception-only. Inference · Catalog judgment · Medium confidence · `openai-software-factory--operating-models-0` Confidence reason: The article documents agent auto-approve for low-risk pull requests in opted-in areas and risk classification that routes high-risk changes to stricter review; the run-time exception path is not described. The harness-engineering post corroborates agent automerge of low-friction pull requests in one repository but describes no risk tiering, so the routing mechanism itself stays single-sourced. Qualifications: - Observation date: 2026-09 Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, line 159 (risk classification; low-risk areas can opt in to an agent that auto-approves pull requests) - Contextualizes · [3] [Harness engineering: leveraging Codex in an agent-first world](https://openai.com/index/harness-engineering/) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-harness-engineering-post/content.md) · Preserved content.md, line 199 (background Codex tasks 'reviewed in under a minute and automerged' in one repository; no risk-tiered routing described) ### Operating model assessment Level 3 for code change -> production rollout on the general path; human attention boundary: work-product-review. Inference · Catalog judgment · High confidence · `openai-software-factory--operating-models-1` Confidence reason: Pipeline step 6 states that a human approves a change to go to production before the deploy agent takes over. Qualifications: - Observation date: 2026-09 Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, lines 165-175 (after a human approves production, an assigned agent handholds the rollout) ### Operating model assessment Unclassified for production alert -> proposed performance fix; human attention boundary: unknown. Inference · Catalog judgment · Unverified confidence · `openai-software-factory--operating-models-2` Confidence reason: The article documents Perf Factory only up to agents that de-duplicate signals, identify real latency regressions, root-cause them, and propose fixes; whether a human reviews, applies, or lands those proposals is not stated. Qualifications: - Observation date: 2026-09 Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, line 186 (Perf Factory proposes fixes) ## Implementation details ### Sandbox unknown Inference · Catalog judgment · Medium confidence · `openai-software-factory--architecture-sandbox` Confidence reason: The article does not document an execution sandbox; unknown does not mean absent. Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, lines 138-186 describe the pipeline without documenting an execution sandbox ### Harness Internal Codex, described as much more advanced than the external product because it is plugged into almost every OpenAI system; ChatGPT Work runs on the same harness Fact · Reported · High confidence · `openai-software-factory--architecture-harness` Confidence reason: A linked first-party source states the claim. Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, line 48 (internal Codex is a lot more advanced than its external counterpart; ChatGPT Work is powered by the Codex harness) - Supports · [5] [zbrock: the Codex app began as an internal prototype](https://news.ycombinator.com/item?id=48435137) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/hn-zbrock-internal-prototype/content.md) · Preserved content.md, line 12 (harness-engineering co-author: 'It was an internal prototype that looked very much like the current Codex app') - Supports · [7] [Introducing the Agents API](https://openai.com/index/introducing-the-agents-api/) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-agents-api-post/content.md) · Preserved content.md, lines 12-14 (OpenAI scaled Codex and ChatGPT for Work on the same harness; the Agents API exposes 'that same harness and infrastructure that powers Codex') ### Model unknown Inference · Catalog judgment · Medium confidence · `openai-software-factory--architecture-model` Confidence reason: The article names no underlying models for the internal harness; unknown does not mean absent. Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · The preserved article names no underlying models for the internal Codex harness ### Interfaces desktop, cli, slack, github, skill Fact · Reported · Medium confidence · `openai-software-factory--architecture-interfaces` Confidence reason: A linked participant or independent source reports the claim. Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, lines 48, 80, and 140-145 (Codex app and CLI, GitHub, Slack, internal skills) ### Tool access Git repositories and GitHub, Slack, Notion, Databricks, Datadog, and internal logs and data sources Fact · Reported · Medium confidence · `openai-software-factory--architecture-tool-access` Confidence reason: A linked participant or independent source reports the claim. Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, lines 140-145 ### Knowledge OpenAI moved its documentation inside the source code; internal Codex skills, some maintained by Codex itself; new engineers are directed to ask Codex during onboarding Fact · Reported · High confidence · `openai-software-factory--architecture-knowledge` Confidence reason: A linked first-party source states the claim. Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, lines 140-147 (documentation inside the source code; internal skills; onboarding) - Supports · [3] [Harness engineering: leveraging Codex in an agent-first world](https://openai.com/index/harness-engineering/) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-harness-engineering-post/content.md) · Preserved content.md, lines 30, 73, and 117 (structured docs/ directory as the system of record; a roughly 100-line AGENTS.md serving as a map; a recurring doc-gardening agent; the initial AGENTS.md was written by Codex) - Supports · [9] [Transcript: 'How OpenAI's Codex Team Uses Their Coding Agent'](https://every.to/podcast/transcript-how-openai-s-codex-team-uses-their-coding-agent) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/every-codex-team-interview/content.md) · Preserved content.md, lines 236, 258, and 274 (automations that keep PRs mergeable by resolving merge conflicts, a random-file bug-finder that runs multiple times a day, and a bot that quietly fixes bugs in recently merged PRs) ### Context management The /goal setting lets an agent work until a goal is complete; threads run for days and spin off other agents Fact · Reported · Medium confidence · `openai-software-factory--architecture-context-mgmt` Confidence reason: A linked participant or independent source reports the claim. Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, lines 52-56 (/goal setting; long-running threads) - Contextualizes · [3] [Harness engineering: leveraging Codex in an agent-first world](https://openai.com/index/harness-engineering/) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-harness-engineering-post/content.md) · Preserved content.md, line 58 (single Codex runs 'upwards of six hours' while humans sleep; hours, not the days the article reports) ## Reported metrics ### Headline claim Almost all OpenAI employees used Codex and ChatGPT Work weekly as of the September 2026 report (self-reported). Metric · Reported · Medium confidence · `openai-software-factory--headline-metric` Confidence reason: Company usage claim relayed by an independent reporter; no method or denominator published. OpenAI's own 2026-09-10 launch post states nearly 100% of teams inside OpenAI, including finance and sales, use ChatGPT Work and Codex (a teams denominator, not employees). Qualifications: - The source does not report the denominator of this figure. - Reported by: OpenAI - Scope: Weekly Codex and ChatGPT Work use across OpenAI employees - Method: Internal token-usage tracking by department, chart sourced to OpenAI - Observation date: 2026-09 Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, line 46 - Supports · [8] [ChatGPT is now a partner for your most ambitious work](https://openai.com/index/chatgpt-for-your-most-ambitious-work/) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-chatgpt-work-post/content.md) · Preserved content.md, line 34 (OpenAI's own launch post: nearly 100% of teams inside OpenAI, including finance and sales, use ChatGPT Work and Codex; teams denominator, OpenAI self-report) - Contextualizes · [9] [Transcript: 'How OpenAI's Codex Team Uses Their Coding Agent'](https://every.to/podcast/transcript-how-openai-s-codex-team-uses-their-coding-agent) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/every-codex-team-interview/content.md) · Preserved content.md, line 486 (Codex lead, February 2026: 'almost everyone technical at the company uses Codex') ### Key observation Non-engineering orgs such as finance, recruitment, and legal went from about 0% to 90% Codex usage within a four-month period (self-reported) Metric · Reported · Medium confidence · `openai-software-factory--key-metrics-0` Confidence reason: Company adoption figures relayed by an independent reporter; the four-month window carries no calendar dates. OpenAI's launch post corroborates finance-org use (month-end close reduced from days to hours) but not the 0%-to-90% trajectory. Qualifications: - The source does not report the denominator of this figure. - Reported by: OpenAI - Scope: Weekly Codex use in non-engineering orgs such as finance, recruitment, and legal - Observation date: 2026-09 Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, line 46 (non-engineering orgs went from ~0% to 90% usage in a four-month period) - Contextualizes · [8] [ChatGPT is now a partner for your most ambitious work](https://openai.com/index/chatgpt-for-your-most-ambitious-work/) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-chatgpt-work-post/content.md) · Preserved content.md, line 37 (OpenAI's finance org uses ChatGPT Work for month-end close and forecasting, reduced from days to hours; corroborates non-engineering adoption, not the 0%-to-90% trajectory) ### Key observation Some build-test-deploy systems saw about a 10x load increase within roughly six months (self-reported) Metric · Reported · Medium confidence · `openai-software-factory--key-metrics-1` Confidence reason: A VP of Engineering load estimate relayed by an independent reporter. Qualifications: - The source does not report the denominator of this figure. - Reported by: OpenAI - Scope: Load on build-test-deploy pipeline systems - Observation date: 2026-09 Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, line 92 (roughly 10x load increase on some systems in about six months) ## Lessons and interpretation ### Lesson Longer-running /goal tasks drove adoption; one long-running agent that spins off other agents reduces the surface a person manages Opinion · Reported · High confidence · `openai-software-factory--lessons-learned-0` Confidence reason: Stated explanation by the Codex desktop lead in the reported interviews. Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, lines 52-56 (Andrew Ambrosino on long-running tasks and agents spinning off agents) ### Lesson Role-specific and team-specific plugins spread adoption beyond a generic coding agent Opinion · Reported · High confidence · `openai-software-factory--lessons-learned-1` Confidence reason: Stated explanation by the Codex desktop lead in the reported interviews. Evidence: - Supports · [1] [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/openai-factory-article/content.md) · Preserved content.md, line 64 (role-specific and team-specific plugins are created and distributed) ## Related implementations - Names this entry as its context: [OpenAI — Sevbot](https://internal-agents.com/agents/openai-sevbot) ## Sources 1. [Inside OpenAI's agentic software factory](https://newsletter.pragmaticengineer.com/p/openai-software-factory) - News · Independent secondary · Evidence - Original URL: - Publisher: The Pragmatic Engineer · Published: 2026-09-15 · Accessed: 2026-09-16 · Last verified: 2026-09-16 - Preserved copy in the repository: 2. [The most important OpenAI announcement you probably missed at DevDay 2025](https://venturebeat.com/infrastructure/the-most-important-openai-announcement-you-probably-missed-at-devday-2025) - News · Independent secondary · Evidence - Original URL: - Publisher: VentureBeat · Published: 2025-10-09 · Accessed: 2026-09-16 · Last verified: 2026-09-16 - Preserved copy in the repository: 3. [Harness engineering: leveraging Codex in an agent-first world](https://openai.com/index/harness-engineering/) - Engineering blog · First party · Evidence - Original URL: - Publisher: OpenAI · Published: 2026-02-11 · Accessed: 2026-09-16 · Last verified: 2026-09-16 - Preserved copy in the repository: 4. [Harness engineering: Leveraging Codex in an agent-first world (discussion)](https://news.ycombinator.com/item?id=48416264) - Hn thread · Community · Commentary - Original URL: - Publisher: Hacker News · Published: 2026-06-05 · Accessed: 2026-09-16 · Last verified: 2026-09-16 - Preserved copy in the repository: 5. [zbrock: the Codex app began as an internal prototype](https://news.ycombinator.com/item?id=48435137) - Hn comment · Direct participant · Evidence - Original URL: - Publisher: Hacker News · Published: 2026-06-07 · Accessed: 2026-09-16 · Last verified: 2026-09-16 - Preserved copy in the repository: 6. [zbrock: many internal teams adopted the same practices](https://news.ycombinator.com/item?id=48435213) - Hn comment · Direct participant · Evidence - Original URL: - Publisher: Hacker News · Published: 2026-06-07 · Accessed: 2026-09-16 · Last verified: 2026-09-16 - Preserved copy in the repository: 7. [Introducing the Agents API](https://openai.com/index/introducing-the-agents-api/) - Corporate article · First party · Evidence - Original URL: - Publisher: OpenAI · Published: 2026-09-10 · Accessed: 2026-09-16 · Last verified: 2026-09-16 - Preserved copy in the repository: 8. [ChatGPT is now a partner for your most ambitious work](https://openai.com/index/chatgpt-for-your-most-ambitious-work/) - Corporate article · First party · Evidence - Original URL: - Publisher: OpenAI · Published: 2026-09-10 · Accessed: 2026-09-16 · Last verified: 2026-09-16 - Preserved copy in the repository: 9. [Transcript: 'How OpenAI's Codex Team Uses Their Coding Agent'](https://every.to/podcast/transcript-how-openai-s-codex-team-uses-their-coding-agent) - Transcript · Independent secondary · Evidence - Original URL: - Publisher: Every · Published: 2026-02-18 · Accessed: 2026-09-16 · Last verified: 2026-09-16 - Preserved copy in the repository: