DEV Community

Cover image for Our one-cent agent budget stopped at 47 cents, and the test still passed
Justin Bartak for Orbyt Labs

Posted on Originally published at orbytlabs.ai AI-assisted

Our one-cent agent budget stopped at 47 cents, and the test still passed

A one-cent ceiling stopped our agent session only after its estimated usage reached $0.478471, nearly 48 times the limit it was supposed to enforce.

  • Both runs ended with error_max_budget_usd, and both records say passed: true. The checker passed on any one of three markers and never required the estimate to stay under the limit.
  • Our own protocol's final paragraph says the controls prove a reservation. Its Deviation 2 says the stop came after the crossing call, and the raw records side with Deviation 2.
  • In a synthetic case, two workers each see $0.40 left on a $1.00 shared limit and each approve a $0.30 action. Both are locally affordable, and total usage can still reach $1.20.

Read the full article: Can Your Agent Outspend Its Limit?

This summary was written with AI assistance (Claude) from the full article at https://www.orbytlabs.ai/blog/ai-agent-budget-limits.

Top comments (0)