The Autonomy Level Slider: Cautious, Balanced, High
The Autonomy Level slider controls how independently the AI works. It’s a scale from 0 to 100 that determines when the AI pauses to ask for your input versus when it proceeds on its own.
Where to find the Autonomy Level slider:
- Open the Dashboard Settings → AI tab
- In the “AI Intelligence” section, you’ll see the “Autonomy Level” slider
- Drag the slider to set your preferred autonomy level (0-100)
The three autonomy ranges:
Cautious (0-49): Maximum Oversight
In Cautious mode, the AI confirms each major step with you before proceeding. It works like a junior developer who asks for approval frequently.
Behavior:
- AI presents a plan and waits for approval
- AI confirms before each phase or major task
- AI asks clarifying questions when ambiguity exists
- AI pauses when encountering unexpected results
- AI reports progress after each completed step
Best for:
- Learning how GameCatalyst works
- Working on critical or unfamiliar features
- Projects where mistakes are costly
- Developers who prefer hands-on control
Balanced (50-79): Independent with Checkpoints
In Balanced mode, the AI works independently but pauses for important decisions, user comments, and divergent situations. It’s like a mid-level developer who knows when to ask for input.
Behavior:
- AI executes approved plans without step-by-step confirmation
- AI pauses when user leaves a comment on a plan or task
- AI asks for guidance when multiple valid approaches exist
- AI reports issues but attempts recovery before asking for help
- AI acknowledges all user comments before advancing
Best for:
- Most developers and most projects
- Balancing speed with oversight
- Collaborative workflows where you provide occasional guidance
- Projects where you trust the AI but want checkpoints
High (80-100): Maximum Automation
In High mode, the AI proceeds independently and only stops for critical blockers. It works like a senior developer who makes decisions autonomously.
Behavior:
- AI creates and executes plans without waiting for approval (if plan approval is off)
- AI makes implementation decisions without asking
- AI only pauses for critical errors or blockers
- AI handles ambiguity by choosing the most reasonable approach
- AI reports completion summaries rather than step-by-step progress
Best for:
- Experienced developers who trust AI assistance
- Rapid prototyping and iteration
- Repetitive tasks where the approach is clear
- Projects where speed is more important than oversight
How autonomy level affects AI behavior:
The autonomy level is injected into the AI’s behavioral directives. When the AI connects, GameCatalyst tells it:
- Cautious: “Confirm each major step with the user before proceeding”
- Balanced: “Work independently but pause for user comments, ambiguity, and divergent decisions”
- High: “Proceed independently, only stopping for critical blockers”
The AI adjusts its behavior accordingly throughout the session.
Combining autonomy with other safety features:
Autonomy level works alongside other safety features:
- Dry-run mode — Still active regardless of autonomy level. High autonomy doesn’t bypass dry-run.
- Protected paths — Still enforced. High autonomy doesn’t grant access to protected directories.
- Permission gates — Still checked. High autonomy doesn’t enable disabled commands.
- Plan approval — Can be combined with any autonomy level for additional control.
Think of autonomy level as controlling the AI’s decision-making independence, not its safety constraints. Safety features remain active at all autonomy levels.
