What Happened
OpenAI launched a research preview of Codex: a cloud-based software engineering agent designed to work on multiple tasks in parallel (e.g., implementing features, fixing bugs, answering codebase questions, proposing pull requests), each in its own sandboxed environment.
Why It Matters
Codex reframed “AI coding” from inline autocomplete toward agentic software engineering: longer-running, multi-step work with execution, iteration, and review artifacts that resemble real development workflows.
Technical Details
- Execution model: Parallel tasks in isolated environments
- Agent outputs: Proposed code changes suitable for review workflows (e.g., PR-style diffs)
- Model note: Announced as powered by
codex-1in the release post