Before anyone picks a model or a framework, there's a more basic question: is this workflow actually ready to be automated? We've watched teams buy the shiniest agent and watch it fail because the workflow behind it was mush. The tool is rarely the problem. Readiness is.
This checklist screens for the things that actually predict whether a deployment sticks. If you score low, that's not a verdict — it's a to-do list, and most of it is afternoon work.
The five signals that predict success
The eight questions boil down to five things, and none of them are about technical pedigree:
Defined work. The task follows the same pattern each time, with clear inputs and a clear finish line. "Make the business better" is not automatable. "Reply to every inbound lead within an hour using this template" absolutely is.
Reachable data. The agent needs to get at the data — CRM, inbox, a sheet — without a forensic exercise. If the information lives in someone's head or three unlinked tools, sorting that out is the first job, not the agent's.
A human on the sensitive step. The best deployments keep a person approving the thing that matters. That's not a weakness; it's what keeps the agent trusted and shippable from week one.
Enough volume. Automation has setup cost. A task you do twice a year isn't worth a build; one you do daily absolutely is.
An owner. Someone has to care that the agent keeps working after launch. No owner means it silently rots, and we've seen that happen more than once.
What a ready workflow looks like
Concrete example: a B2B team gets thirty demo requests a week. The task is defined — qualify, enrich the CRM, send a booked-meeting link — the data's in HubSpot and Gmail, a rep approves the send, there's clear daily volume, and the founder owns it. That scores 8/8 and we'd build it the same week.
Contrast that with "improve customer happiness." No definition, no clear finish line, no owner. It might score 2/8 — and the right move there isn't to automate, it's to go define the workflow first. The checklist is really just forcing that honesty early. Working a whole process down into steps an agent can own is covered inautomating a business process with AI agents.
If you scored low
Low scores read like bad news. They rarely are. Most gaps are about clarity, not capability — tidy the data, write down the actual steps, name an owner. Those are afternoons of work, not quarters of re-architecture. Half of a "we're not ready" verdict is really "we never wrote the process down," and writing it down is most of the work.
When you should NOT automate
Worth saying out loud: some things stay human. A task where a mistake is rare but catastrophic. A relationship-defining moment with a key account. Anything where you'd be uncomfortable explaining the automation to the person affected. The point is to take the repetitive weight off people, not to remove the human from the parts that need one — and if a task falls in that bucket, leave it. We'll say so.
A high score also does not mean you need an agent specifically. If the work is answering the same questions in one channel, the difference between an AI agent and a chatbot is the distinction to settle first. If it does need an agent, how long a production AI agent takes to build sets the expectation for what happens after a good score.