A horizontal spectrum with three stops, left to right. Scripted: every step hardcoded; predictable; easy to control. Semi-autonomous, highlighted as where most real-world agents sit: picks from a defined toolkit within set guardrails. Highly autonomous: decides what to search, fetch, and revise; may write and run new functions; powerful but harder to control. The footer principle reads: more autonomy means more need for evaluation, permissions, and guardrails.

1. Scripted

   Every step hardcoded. Predictable. Easy to control.

2. Most real-world agents sit here
   Semi-autonomous

   Picks from a defined toolkit within set guardrails.

3. Highly autonomous

   Decides what to search, fetch, and revise. May write and run new functions. Powerful but harder to control.

More autonomy → more need for evaluation, permissions, and guardrails.