Rate limits are how every meaningful API protects itself from being overwhelmed. The CRM allows 60 calls per minute per token. The calendar service allows 1000 requests per hour per account. Cross the limit and the API starts returning 429 errors, dropping the calls the agent is depending on.
For a CX-AI agent, rate limits are a real production concern, not a corner case. A call spike (Monday morning at the clinic, the day after a storm at the dispatcher) drives the agent to make hundreds of CRM calls in a short window, exactly the kind of traffic that trips rate limits.
Production engineering handles this with adaptive throttling, request batching, and read caching. A vendor that has not thought about rate limits will tell you "the integration just works" until the day it does not, usually during the customer's busiest hour.
Vorel adapters maintain per-CRM rate-limit budgets, with adaptive throttling and read caching to keep the agent under quota during traffic spikes. Quota breaches show up in monitoring before they affect calls.