GitHub letting developers choose between coding agents is more important than it first looks. The model matters, but the place where the work happens matters too. If an agent can see the issue, branch, pull request, review comments, permissions, and history, it starts to become part of the development platform rather than a separate tool pasted onto the side.
GitHub announced that Claude and Codex are available in public preview on GitHub, and later added model selection for those third-party agents. That means a task can stay attached to the GitHub workflow while the underlying agent choice becomes more flexible.
For teams, this is a different decision from “which chatbot do we like?” It is about governance, auditability, access control, billing, review flow, and how much context the agent gets by default. An agent inside GitHub can be convenient because it starts near the work. It can also be risky if teams treat that convenience as a substitute for review.
My guess is that most engineering teams will end up with more than one agent available, but fewer places where agents are allowed to act. That feels like the right trade. Let developers pick the tool that fits the job, but keep the actual work inside systems where review, history, and permissions are already understood.