Hiring an AI Automation Agency: 9 Questions That Expose the Pretenders

The AI agency gold rush has flooded the market with rebranded dev shops. Nine questions that separate operators from opportunists — and the answers that should end the meeting.

Every gold rush mints more shovel-sellers than miners. AI automation is no different — which means your vendor call is now a filtering exercise. These nine questions do the filtering fast. Yes, clients should ask us these too.

The nine questions

• 1. “What do you run in your own business?” — Operators show dashboards. Pretenders show slides.
• 2. “Walk me through a deployment that failed.” — Anyone shipping real systems has one. No failures means no deployments.
• 3. “What would you NOT automate in my business?” — Judgment is the product. “Everything!” is the wrong answer.
• 4. “Where does a human approve before the system acts?” — No approval gates near money = amateur hour.
• 5. “What happens when the model provider has an outage?” — Real systems have fallbacks. Demos don’t.
• 6. “Who owns the workflows, prompts, and data when we part ways?” — The only acceptable answer: you do.
• 7. “What’s the monthly run cost, separate from your fee?” — Vagueness here predicts vagueness everywhere.
• 8. “How fast is the first process live?” — Weeks. If the answer is quarters, you’re funding their learning.
• 9. “Show me the monitoring.” — Systems without observability fail silently, and you find out from a customer.

You’re not hiring intelligence — the models supply that. You’re hiring judgment about where intelligence should touch your business.

The meta-signal

Notice which questions make a vendor uncomfortable. Discomfort at #1 and #2 means they’ve never operated. Discomfort at #6 and #7 means the business model is lock-in. Comfort across all nine — with specifics — is what competence sounds like.