When APIs are consumed by agents, not humans

As APIs become consumed by agents instead of mostly humans, traffic patterns may shift from bounded UI-driven interactions to high-fanout, recursive, machine-speed workflows.

This could materially change scaling assumptions around API gateways, rate limits, caching, database hot partitions, write amplification, and read-heavy scan patterns.

The core primitives remain the same, but systems may need more agent-aware APIs, quotas, batching, async execution, and data architecture optimized for machine consumption.