Agents API enters public beta
A managed harness is an architectural choice, not simply a model-name replacement.
Independent guide · The OpenAI ecosystem
Understand OpenAI.
Beyond the headlines.
What changed, who it affects, and what to do next. Find your way through ChatGPT, Codex, APIs and business tools—with official sources and evidence you can inspect.
Field guide 001Evidence cutoff 10.09.26
2015
2026
01What could
the systems do?
the systems do?
02What paid for
their development?
their development?
03Who could constrain
the decision-makers?
the decision-makers?
19 chapters71 references1 unresolved story
Change desk / 11 September 2026
The change is only
half the story.
Project API keys can have expiration dates
A credential can stop working even when application code is unchanged.
The standalone Codex MCP server has been removed
Launching the removed command can break an integration; ordinary external MCP connections are a different case.
Start with the product map
More than one product.
One place to get oriented.
The useful question is not whether OpenAI is simply a triumph or a betrayal.
It is how technical progress kept changing the resources, institutions, and incentives required to pursue its public mission.
01 Reading the record
Three questions keep the story honest.
Announcements show what institutions said. Research shows what systems demonstrated. Public records and independent reporting show where claims met consequences.
Capability
What did a model or system demonstrably do—and under what conditions?
Benchmarks are not deploymentInfrastructure
Which capital, computing, partnerships, and distribution made it possible?
Scale has institutional costsGovernance
Who held formal authority, practical leverage, and public accountability?
Mission is not control02 The arc
Five eras.
One accumulating tension.
Follow the organization from nonprofit laboratory to mass-market platform and a redesigned public-benefit structure.
- I 2015—2018 Mission & shared research A nonprofit laboratory, practical safety work, and the first generative pretraining experiments.
- II 2019—2020 Capital & scale The capped-profit structure, Microsoft partnership, GPT-3, and hosted API access.
- III 2021—2023 Products & mass adoption Images, code, speech, ChatGPT, GPT-4, and a rapidly commercializing platform.
- IV 2023—2024 Crisis & consequence Board upheaval, departures, privacy, copyright, labor, and multimodal systems.
- V 2025—2026 Agents & institutional redesign Delegated computer work, infrastructure expansion, and another answer to the governance problem.
The cornerstone essay
Public mission, commercial scale, and the struggle to govern artificial intelligence.
The complete web edition separates technical achievement from institutional claim, and declared intent from demonstrated outcome.
Read all 19 chapters“Useful capabilities and legitimate control are related questions. They are not the same question.”
Editorial conclusion
03 What comes next
Updates should reduce uncertainty—not merely repeat announcements.
The change desk explains what moved. The product directory maps what exists. Our file lab shows downloadable Codex-assisted examples, automated checks, and the limits those checks do not settle.
Inspect the file evidence