A practical filter for separating genuine agent opportunities from expensive demos.
Agent demos are impressive and frequently automate something nobody needed automated. A short filter separates the workflows worth building from the ones that only look good in a video.
Volume times duration
The first question is how many times a week this happens and how long it takes each time. A task done twice a month is not worth an agent regardless of how tedious it is. A five-minute task done two hundred times a week is worth serious engineering.
Is the outcome checkable
An agent whose work cannot be verified cannot be trusted or improved. If a human can look at the result and say quickly whether it is right, you can build evaluation, catch regressions and let it run. If correctness is a matter of opinion, you have a research project rather than an automation.
What does a mistake cost
Drafting an internal summary and issuing a refund are not the same risk. Start where errors are cheap and reversible, prove reliability, then move toward higher-stakes work with approval gates in place.
Is the process stable
Automating a workflow that changes every quarter means rebuilding it every quarter. Stability is worth more than complexity here.
Score candidate workflows on those four points before writing any code. The best opportunity is usually duller than the one that generated the enthusiasm.
Want this for your business?
Let's talk about how we can help you build and grow.
