Unbounded or unmonitored LLM spend driven by loops, retries, oversized contexts, or misconfigured agents.
Below: real production examples of runaway agent costs, the root causes, vendor-neutral prevention techniques, and detection signals to monitor.
What it actually looks like in production
- Support agent retried a failing tool 47 times: $1,200 in one conversation
- Research agent with max_iterations=20: $14 per run × 800/day = $336k/month
- Multi-agent system ping-pong: $40 per conversation due to weak termination
Why it happens
- Loops without terminal conditions
- Tool-call retries on transient errors without backoff/caps
- High max_iterations defaults
- Multi-agent conversations without convergence criteria
- Wrong model for the task
- Unbounded context history growth
How to prevent it (vendor-neutral)
1. Set hard iteration limits per agent
2. Set per-run cost ceilings
3. Set per-user and per-tenant daily caps
4. Set wall-clock timeouts
5. Match model tier to task complexity
6. Use prompt caching where available
7. Truncate or summarize old history
8. Differentiate transient vs. semantic retry semantics
How Prefactor helps detect and prevent it
Prefactor sits at the agent runtime and contributes specifically:
- Runtime guardrails that flag or block matching patterns before they land
- Continuous eval suites that catch quality regressions on every change
- Tamper-evident logs of every incident and response action
- Per-agent anomaly alerts on the signals listed below
Detection — what to monitor
- p95/p99 per-run cost outliers
- Runs hitting max_iterations regularly
- Outlier tenant or user spend
- Average context size growing over time
Response — what to do when it happens
Immediate (minutes): confirm the incident from the trace; pause the affected agent if active harm possible; hotfix the trigger.
Short-term (hours): add the failure case to the eval suite; patch the root cause; redeploy with regression validation.
Medium-term (days): root cause analysis; tighten guardrails or controls; document the incident for post-mortem and audit.
FAQ
Can runaway agent costs be eliminated entirely? Usually no — reduce frequency and severity dramatically, and contain blast radius. Aim for low, detected, and contained.
How often should we test for this? Continuously, with every change. Every reported incident becomes a test case.
Can Prefactor detect this in real time? Yes for many variants — guardrails run in-line with sub-second latency.
Related
See Prefactor in action
[Get started free →] [Book a demo →]