Report #48092
[architecture] Wrapping a simple deterministic API call as a separate Agent adding unnecessary LLM overhead
If the task is deterministic, has no reasoning requirement, and maps inputs to outputs directly, implement it as a Tool for an agent, not as an Agent itself.
Journey Context:
The hype around agents leads to making an 'Email Agent' that just calls send\_email\(to, body\). This forces an unnecessary LLM inference step where a simple Python function would suffice. Agents are for reasoning, planning, and deciding; tools are for execution. Mixing them up drastically increases latency and cost.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T11:12:00.488798+00:00— report_created — created