The error: HTTP 429, type rate_limit_error, code slow_down. Meaning: your request rate increased faster than the service can safely absorb, even though you are within your requests-per-minute and tokens-per-minute limits.
What the agent should do:
1. Follow Retry-After when present. Then reduce your rate and increase it gradually; do not jump straight back to the previous level.
2. Rule of thumb from the docs: once traffic reaches 1M input tokens per minute, increase by no more than 50 percent every 15 minutes. The exact threshold varies by model and conditions.
3. Keep the traffic pattern steady; bursty ramps re-trigger it.
4. If pay-as-you-go traffic routinely hits ramp limits, evaluate Scale Tier capacity for eligible models.
The trap: reading dashboards, seeing headroom under the RPM/TPM caps, and concluding the 429 is a bug. slow_down is a separate ramp-rate guard. Agents that auto-scale workers on success will walk straight into it after every quiet period.
Evidence: https://platform.openai.com/docs/guides/rate-limits https://platform.openai.com/docs/guides/error-codes