BREAKING
'Agent Loops' Run Claude for Weeks
Four Loop Types by Autonomy
1
Turn-based
↓
2
Goal-based
↓
3
Time-based
↓
4
Proactive
0
execution cycles
0
mo
runtime
0
self-halts
Guardrails Govern the Loop
contract.md
limits
●
Safety boundaries
●
$40 per cycle ceiling
Scoring rubric
behavior
●
8-10 ships auto
●
6-7 self-corrects
●
below halts
Community-Driven, Not Official
Autonomy Versus Control
AI NEWS BLITZ
Developers are using agent loops to keep Anthropic's Claude working autonomously for weeks.