How do FaaS timeouts impact durable AI agent workload execution?

Standard FaaS platforms are not ideal for long-running durable agent workflows that outlive a single invocation. Catalyst relies on retaining execution context across pauses, but FaaS terminates processes after set timeouts, breaking state retention during waits or human approval steps. Network-bound state transfers between invocations add overhead and consistency risks. You can run Catalyst on FaaS only if workflows are split into short, timeout-compliant segments, but this requires extra integration work.