Copilot Automations add budgets to agent workflows

An automation is only useful when a team can predict when it runs, what it can touch, and how much it can consume. GitHub's October update for Copilot Automations adds those controls around the prompt: external API triggers, more GitHub events, chained runs, new tools, and AI credit limits.

The changelog says automations can now respond to more issue, pull request, discussion, and security events. Teams can connect custom MCP servers and set limits per run, day, week, or month. A workflow can also trigger another automation after success or failure.

Budget is part of the completion contract

The new controls make an agent workflow easier to place inside normal operations, but a limit is not a quality check. A run that stops at its credit ceiling may still leave a partial diagnosis, a misleading label, or a task that nobody owns. The workflow needs an explicit fallback and an accountable reviewer.

The first question for an automation is not what it can do. It is what happens when it runs out of money halfway through.

Start with one trigger, one bounded tool set, and a read-only output. Record the evidence a human needs to accept the result, cap the spend, and make the failure path visible before adding chained actions or write access.