Agentic Workflows: generally available in Actions
GitHub’s October 28 announcement makes Agentic Workflows generally available on github.com for GitHub Free, Team, and Enterprise. Using its default Copilot CLI engine still requires a Copilot license; GitHub Free availability does not make every model-powered run free. We have not established the announcement’s first public hour. [1]
These workflows turn natural-language Markdown definitions into GitHub Actions YAML. GitHub describes read-only agent permissions by default, a sandbox and firewall, validated outputs, and a threat-detection job. Those are described safeguards, not an assurance that generated changes are correct or risk-free. [1]
Shared Copilot Automations: a distinct rollout
Separately, GitHub says shared Copilot Automations are rolling out to Pro, Pro+, Max, Business, and Enterprise users, with availability varying during rollout. The Automations guide restricts them to private or internal repositories and requires cloud-agent access. [2][3]
For a shared automation, repository readers can see it and writers can edit it. It belongs to the repository, acts as Copilot, and bills AI credits to the repository owner rather than its creator. Organization-owned repositories need the relevant organization-billed cloud-agent policy enabled. [2]
The rollout-linked guide says Actions minutes are billed separately to the repository owner. Its daily, weekly, and monthly AI-credit limits prevent new runs; they do not stop an active run. A per-run limit serves that different purpose. [3]
Check the type shown in your interface
Our source fetches returned both an older personal-only Automations guide and a newer rollout-linked version. The October 28 changelog supports the shared-ownership rules above. Confirm that the automation itself is Shared. Existing personal automations retain their personal identity and billing rules; the type cannot be changed after creation. [2][3]
Our take
For teams considering recurring AI work, agree who can change the prompt and tools, who pays, who reviews the output, and how a run can be stopped. Treat model credits and runner charges separately. Start with a small repository task and inspect its actions before expanding access.
We reviewed public sources; we have not created or tested either system.