GitHub Copilot cloud agent adds per-task reasoning-level control

GitHub shipped reasoning-level selection for Copilot cloud agent on August 3, 2026: when delegating a task you now pick how much the model reasons before responding, trading answer quality on complex problems against token/credit spend. Available on all paid Copilot plans that include the cloud agent.

What it means Reasoning effort is becoming a first-class, user-tunable cost dial across agent products — budget-vs-quality control is now part of delegating work.