OpenAI hires Git AI founders as agent tooling tightens and self-hosted coding lands

OpenAI absorbs Git AI to ground agent accounting in open-source provenance

OpenAI has hired the founders of Git AI, an open-source tool that tracks how much code AI writes and measures the performance and cost of coding agents, the New Stack reported on September 13, 2026. Git AI's stated mission is unchanged under the larger platform, and the move puts attribution, cost and performance telemetry for coding agents inside the same vendor shipping those agents. For practitioners, the practical change is that the accounting layer for "how much of this PR is mine versus the model's" stops being a community project and becomes a first-party signal from an agent vendor with its own incentives.

Teams that relied on Git AI as a neutral check should plan for the project's governance to shift with the team.

Self-hosted AI coding moves from experiment to product

Coder introduced a self-hosted AI coding product on September 13, 2026, according to SD Times, letting enterprises run coding agents inside their own infrastructure rather than routing source code through external model APIs. The shift matters for teams blocked by data-residency, IP or vendor-risk reviews: the unit of deployment is no longer "an IDE plugin" but "an internal coding service" with its own auth, logging and cost controls. For developers, the immediate workflow impact is that code completion, refactor and review suggestions can now run against internal repositories without the prompt leaving the corporate network, at the cost of standing up GPU capacity and an evaluation harness internally.

React Developer Tools ships across Chromium browsers and Firefox

Alongside the agent tooling news, React announced a new release of React Developer Tools that are now available in Chrome, Firefox and Chromium Edge, with performance gains over the previous build, per the same September 13, 2026 SD Times report. The cross-browser parity removes one of the recurring friction points for React developers who profile applications from a non-Chromium default browser. Practitioners debugging component trees in Firefox, or in Edge variants that track Chromium, should expect a comparable experience to Chrome DevTools for the React-specific panels.

A coding agent runs unchecked for 30 hours

A software company founder went viral this week after posting on social media describing how an AI coding agent threw his business into chaos for 30 hours, ABC News reported on September 13, 2026. The founder remains publicly optimistic about the technology. The episode is a concrete reminder that persistent coding agents, the same class of tool that OpenAI and Coder are scaling, need explicit blast-radius limits: repository scope, branch protection, dry-run mode, and kill switches on long-running tasks. Teams adopting any of the products above should treat the failure mode as a known category, not a novel risk.

What to check next

Three concrete follow-ups for September 13, 2026: first, watch the Git AI repository for a governance update from its new employer and decide whether to keep it as your attribution source of record; second, if self-hosted coding is on your roadmap, ask Coder for a reference architecture covering GPU sizing, model selection and audit logging before you pilot; third, audit any long-running agent jobs you have scheduled for hard time limits and repository allow-lists, in light of the 30-hour failure reported this week.

For day-to-day browser debugging, confirm your What Browser Am I Using check still matches the browser you launched React Developer Tools from, since the new release ships to Chrome, Firefox and Chromium Edge in parallel.