Eight questions to ask before you hire anyone to build an AI agent or automation for your business, and the answers that should make you walk away.
The AI automation market attracts a wide range of skill levels behind similar-looking pitches: a demo video, a few client logos, a promise to "save you hours." Ask these eight questions before you sign anything. The answers separate people who build production systems from people who connect a few no-code modules and call it an AI agent.
You should get a written description of the system before work starts: what triggers it, what it checks, what it does, and what counts as success. If the proposal describes an outcome ("save you 10 hours a week") without describing the mechanism, push back and ask for the mechanism.
Walk away if the answer stays vague after you ask twice.
You should. Ask for the code, the credentials, and setup documentation to be handed over at project close, in a format you can read without the person who built it. Some partners build inside their own accounts and treat access as leverage to keep you paying.
Walk away if ownership isn't addressed in writing before the project starts.
Models get updated, deprecated, or re-priced. APIs change their response format without much warning. A partner who's built production systems before has a plan for this: version pinning, monitoring, a process for testing an update before it goes live. A partner who hasn't usually hasn't thought about it.
Walk away if the answer is "that doesn't really happen."
If the system touches customer data, ask where that data goes, which third-party APIs see it, and whether it's used to train anything beyond your own system. For regulated data (health, financial, or anything under GDPR or similar law) ask for this in writing, not a verbal reassurance.
Walk away if the question makes the person uncomfortable instead of specific.
A demo proves a tool can be made to work once. A system that's been running in production for six months, handling real volume, proves it holds up. Ask for a reference you can contact, not a screenshot.
Walk away if every example is a prototype or a personal project.
Every real agent hits cases the data doesn't clearly answer: a customer record with two different email addresses, an invoice total that doesn't match the line items. A well-built agent flags that case for a person instead of guessing. Ask what that fallback looks like and where it gets logged.
Walk away if the answer is that it always picks something.
A fixed scope should come with a fixed price, agreed before work starts. Ask directly what happens if you request something outside that scope once the project is underway. A clear answer names the process for pricing the addition. A vague one gets folded into a bigger invoice at the end with no warning.
Walk away if the pricing conversation stays abstract past the second meeting.
Some projects end with full documentation and a clean handoff, no ongoing relationship required. Others need a retainer because the system depends on an upstream API that changes or a data source that shifts. Either is reasonable. What isn't reasonable is finding out which one you signed up for only after something breaks.
Walk away if post-launch support was never discussed at all.
A pattern worth noticing: every question above has the same shape. A partner who has built and maintained production systems answers with specifics, because they've hit the problem before. A partner who hasn't answers in reassurances. Listen for the difference.
You should. A partner who hands over documented code and credentials at the end of the project leaves you free to maintain it yourself or move to someone else. A partner who keeps the workflow inside their own account creates a dependency you didn't agree to.
It should stop and flag the case for a person, not guess. An agent that always picks an action, including when the data is ambiguous or missing, will eventually take the wrong action on something that mattered, like a refund or a customer record.
Fixed scope should mean a fixed price, agreed before work starts. If you ask for something beyond that scope mid-project, the partner should name the added cost before building it, not fold it into a larger invoice at the end.
Ask us directly, before you decide anything. We'd rather answer up front than have you find out the hard way three months in.
Start a Conversation →