Getting Startedbeginnercurrenttested against paperclip 0.9
Where to go after your first successful run
Once one agent works, grow the system by adding one new variable at a time.
Expand carefully
After the first successful run, the right move is not “add five more agents.”
Add one thing at a time:
- one new task type
- one new specialist agent
- one new governance rule
- one budget limit
Sequence that works
Use this order:
- repeat one successful task
- add a second task type
- add a second agent only if role separation is obvious
- add review or approval only after output volume creates risk
Why slow growth wins
Paperclip gets powerful fast. That also means you can hide a broken mental model under lots of moving parts. Slow growth keeps the failure surface small.