Agents vs workflows: stop calling everything an agent
A clearer mental model for when to use an agent loop, a workflow graph, or a single LLM call.
If your 'agent' has a fixed five-step plan that runs in the same order every time, it's not an agent. It's a workflow. Calling it an agent buys you nothing and costs you debuggability, latency and money.
Real agents are loops where the model decides the next tool call based on what it just learned. They are powerful when the input space is too large to enumerate and the model genuinely needs to explore. They are overkill for anything that fits in a Make scenario.
The honest rule of thumb: start with a single LLM call. Promote to a workflow when you have more than three deterministic steps. Promote to an agent only when the next step truly cannot be known in advance.