Codex turns ChatGPT into developer workflow

This is a distribution land grab, not just a feature launch. By turning ChatGPT into a place where code can be planned, written, tested, and packaged into pull requests, OpenAI moves up the stack from selling models to owning the full developer workflow, where daily usage is stickier and far more valuable than one off chat prompts. That puts it into direct competition with IDE first tools like Cursor, model led coding products like Claude Code, and autonomous engineer products like Devin.1 2 3 4 5

• Cursor showed how big the prize is. It grew from $1M ARR to $100M ARR in 12 months, then to $200M ARR by March 2025, by living inside the code editor and helping with multi file edits, debugging, and agentic work. Codex is OpenAI’s bid to capture that usage inside ChatGPT instead of leaving it to an external IDE.3 4
• Anthropic proved coding can become a major revenue line for a model lab. Claude Code hit an estimated $400M annualized revenue in July 2025 and was described as generating 8% of Anthropic’s total revenue, showing that software engineering is one of the few AI use cases with heavy repeat usage and clear willingness to pay.2 5
• The category is also widening beyond engineers. Research on Codex automations and Claude Cowork shows the same agent pattern being applied to product, marketing, and ops work. Once an agent can inspect files, run tools, and complete multi step tasks, coding is just the first high value workflow, not the last one.1 2

The next phase is convergence around full work operating systems. Coding agents are becoming the proving ground where labs learn to manage long running tasks, tool use, and parallel workflows, then reuse that machinery across every other white collar job that can be broken into files, apps, approvals, and outputs.1 2