We've taken over projects from agencies before. In every case, the client paid a significant amount of money for something that either didn't work, couldn't be maintained, or was built on a platform that locked them in completely.
This post is a buyer's guide. Use it before you sign anything.
The problem with the AI agency market right now
The barrier to calling yourself an AI agency is zero. Anyone can wrap GPT-4 in a chat interface, add their logo, charge $10,000, and call it a bespoke AI solution. Most clients can't tell the difference until it breaks.
The tells are subtle if you don't know what to look for. They're obvious once you do.
Questions to ask before hiring
"Can I see the system prompt?"
If an agency built you a chatbot and won't show you the system prompt, that's a red flag. It's your product. You should own and understand every component. An agency that hides this is either covering up how simple it is or creating lock-in.
"What happens if I want to move this to a different provider?"
The answer should be: you take the code and go. If they describe a migration process that involves their platform, their API keys, or their proprietary system, you're renting, not owning.
"Who actually builds this?"
In most agencies, the person who sells the project is not the person who builds it. Ask specifically. If the answer is vague — "our team of engineers" — ask for the names and backgrounds of the people on your project. You're entitled to know.
"What does failure look like and what happens next?"
A good agency has a clear answer: monitoring alerts, defined SLAs, a process for handling edge cases. A bad agency looks confused by the question.
"Can you show me something you've built that's live in production?"
Demos are easy. Production deployments are hard. Ask for a case study with real numbers — not a slide with a logo on it.
Red flags
They lead with the technology, not the problem. Any agency that opens with "we use the latest GPT-4o model with RAG and vector embeddings" before understanding your business is selling, not solving.
The proposal arrived in 24 hours. A real scoping exercise takes time. A proposal that lands the next morning was written before they understood your situation.
They guarantee results. No one can guarantee AI performance metrics. Anyone who does is either lying or defining the metric in a way that's easy to hit and meaningless.
Everything is custom, nothing is documented. You should receive documentation of your system at handoff. If an agency can't explain what they built in plain English, neither will you when something breaks at 2am.
No fixed price. Time-and-materials billing for AI projects is a trap. Scope should be agreed before work starts. Surprises should be the exception, not the business model.
What good looks like
A good AI agency spends more time asking questions than pitching. They tell you when AI isn't the right solution. They give you a fixed quote before starting. They hand over everything — code, credentials, documentation — at the end.
They're also honest about what they've built before and what they haven't. No agency has done everything. The ones worth hiring know exactly what they're good at and tell you when something is outside that.
We turn down most inquiries. Not because we're busy, but because the fit has to be right before we'll put our name on something.
If you want a second opinion on a proposal you've received, email us. We'll tell you honestly what we think.