A lot of new AI agencies have appeared over the past couple of years. Some are excellent. Others make big claims and have very little to show for them.
As a business owner, it can be hard to tell the difference. Every firm uses the same words. These are the questions I’d ask if I were hiring someone to build AI software for my business.
1. Can you show me working software you’ve built?
Ask for something real, not a deck or a concept video. Better still, ask whether they’ll build a small working piece on your own data before you commit to the full project. We do that because it’s the fastest way for both sides to learn (how our proof of concept works).
2. Who will actually build it?
At some firms, the person who sells the project never touches it. Ask who writes the code and whether you’ll talk with them directly.
3. What’s your engineering background?
AI has made it easier to generate code. It hasn’t made it easier to know whether that code is right. Anthropic’s research on Claude Code found that people with more expertise in the task got verified success far more often than novices (Anthropic). Experience still matters. We cover what that study found in more detail.
4. How do you handle what happens after the demo?
Ask about permissions, backups, error handling, monitoring, and failures. If the firm doesn’t bring these up on its own, be cautious.
5. Where does AI make decisions, and where doesn’t it?
A good answer separates the work. AI should handle judgment calls on messy input, while exact calculations and record matching run in code. Expensive or irreversible actions should wait for a person.
6. Where does my data go?
Find out which AI services your data passes through. For sensitive data, ask whether processing can stay inside your environment.
7. Who owns the code?
You should own it, and you should be able to move hosting or maintenance to another provider. That’s our standard (About Hello Spark).
8. How is pricing structured?
Hourly billing for an open-ended AI project can grow quickly. Ask for a fixed quote tied to a defined scope, and ask how changes are handled.
9. Will you tell me if I shouldn’t build this?
A trustworthy partner will tell you when existing software already does the job or when a build won’t pay for itself.
10. What happens after launch?
Ask about support, monitoring, and improvements. Software that runs a business keeps changing, and you need to know who will be there when it does.
Red flags
Be careful with firms that promise AI will replace whole teams, can’t explain how they test their work, or get vague when you ask about security. Good AI software comes from people who care about the details and have done this kind of work before.