GitHub Turns Copilot's Reasoning Effort Into a Credit Dial

GitHub gave developers a per-task dial for Copilot cloud agent reasoning level, with higher settings burning more credits, per its August 3 changelog.

That choice now sits with whoever kicks off the task, across Pro, Pro+, Business, Enterprise, and Max tiers, turning compute spend into a per-developer bet. Separate research on prompt waste found that effort cues like this one can cost 1.6 to 7.4 times more in compute without improving success rates.

Comment-triggered Copilot automations shipped in the same changelog entry, another way to start agent work without anyone at the keyboard.

The changelog doesn't say if GitHub tested its own reasoning levels against the waste finding. It also doesn't say whether the top setting actually closes more tasks than the lowest one on the same repo.

Exposing the dial is the right instinct. Credits are real money, and a default nobody can see or adjust is worse than one you can tune. But a knob without data on what it buys isn't much better than a guess with a price tag.

What to watch

Whether GitHub, or anyone else, publishes task-completion rates broken out by reasoning level. Until that shows up, "customize your reasoning level" reads as "customize how much you overpay."