Step 1: Task submitted
You describe the task via web UI, GitHub issue (labeled 'jules'), CLI, or REST API.
Step 2: Fresh VM provisioned
An isolated Cloud VM is created. Repository is cloned, AGENTS.md is read, and dependencies are installed.
Step 3: Plan generated and reviewed
Jules creates a step by step plan. The planning critic validates feasibility and completeness before execution begins.
Step 4: Code written and tested
Jules writes code, modifies files, runs commands, and uses the built in browser to verify web application changes.
Step 5: Critic agent reviews
An adversarial code reviewer checks for bugs, logic errors, security issues, and style violations automatically.
Step 6: Pull request created
Changes are committed, pushed to a new branch, and a PR is opened on GitHub with a detailed description and reasoning.
0K+
Public commits during beta
0 tasks
Free daily limit
0 concurrent
Maximum on Ultra tier
Writing effective task descriptions for Jules
Jules works best with specific, scoped tasks. Instead of "improve the codebase," try "Upgrade the React Router dependency from v5 to v6 and update all affected components to use the new API." Include context about what the task should accomplish and any constraints. For recurring patterns, put the instructions in AGENTS.md so you do not need to repeat them in every task.
How the planning critic works
Before Jules writes a single line of code, the planning critic reviews the proposed approach. It checks whether the plan addresses the full scope of the task, identifies potential risks (like breaking changes or missing edge cases), and flags unrealistic assumptions. If the planning critic finds issues, Jules revises the plan before proceeding. This two stage approach (plan, then execute) significantly reduces wasted work and produces higher quality results.