Decide whether this should be an agent at all

I'm considering building an agent for: {{describe the task}}. Volume: {{how often}}. Current cost: {{time or money}}. Consequence of a mistake: {{describe}}.

Assess honestly:
1. How much of this is deterministic? If most steps are rules, say so — a script or a workflow tool will be cheaper and more reliable than an agent.
2. Where is real judgement required, and is it judgement a model is actually good at?
3. What's the cost of a wrong action here, and is it reversible? If irreversible, what must be gated behind human approval?
4. Compare four options on reliability, build cost, running cost, and maintenance: plain script, workflow automation, single LLM call with structured output, and a multi-step agent.
5. Recommend one, and say what would have to change for the next option up to be justified.
6. If an agent is right, name the smallest version worth building and what it should refuse to do in v1.
7. Name the failure that would make me switch it off, and how I'd detect it.

Default to the simplest thing that works. Say so if that's a script.