When Not to Use Agents

Agents are the most over-hyped pattern in AI engineering right now. Don't get me wrong — they're genuinely powerful for certain problems. But I've seen too many teams reach for agents when a simple chain or even a hard-coded pipeline would be better.

Here's my decision framework. Use agents when: the task requires dynamic tool selection based on intermediate results, the number of steps isn't known in advance, and the cost of agent overhead (latency, tokens, complexity) is justified by the flexibility gained.

Use a DAG (directed acyclic graph) when: you know the steps in advance but need conditional branching. Use a simple chain when: the steps are linear and predictable. Use a hard-coded pipeline when: the logic is well-understood and deterministic.

The hidden cost of agents is debugging. When an agent makes a bad decision three steps into a five-step plan, tracing the root cause is significantly harder than debugging a linear pipeline. In production, debuggability is a feature.

Start simple. You can always add agency later. You can't easily remove it.