answers
What should I ask before hiring an AI agent developer or agency?
Before hiring an AI agent developer or agency, ask ten questions: can you show an agent in production, what will it cost to run each month, whose accounts will it run in, what do I receive at handoff, and six more. A table of good answers and red flags.
Before hiring an AI agent developer or agency, ask ten questions, and start with these four: can you show me an agent you built that is running in production today, what will it cost to run each month, whose accounts will it run in, and what exactly do I receive at handoff? The answers separate providers who have shipped working agents from those who have only built demos.
Use the table below on every call, and give each provider the same one-sentence description of your workflow so the answers are comparable.
Updated 2026-09-24 by Esteban Valencia, EV Advisory. Written for founder-led businesses in Canada and the United States.
Which questions matter most?
| Question | Good answer | Red flag |
|---|---|---|
| 1. Can you show me an agent you built that is running in production today? | A live example, with the owner’s permission, and what it measurably does | Only demos, or “all our work is confidential” |
| 2. What will it cost to run each month, not just to build? | A written estimate of model usage, tool subscriptions, and support | “It depends” with no range |
| 3. Whose accounts will the agent run in? | Yours, with their access revocable | Theirs |
| 4. What do I receive at handoff? | A runbook, an access list, the configuration, and a walkthrough | “We will look after it” |
| 5. How will we know it worked? | A baseline measured before the build and the same measure after | No measure, or “you will feel the difference” |
| 6. What should stay human in this workflow? | A specific answer about review and escalation points | “The agent can do all of it” |
| 7. How do you handle personal information? | Minimised data, clear data-use terms, and who can see what | “Privacy is not an issue with this tool” |
| 8. What happens when a connected tool changes and the agent breaks? | A monitoring and support plan with response times | Silence |
| 9. Who exactly will build and support it? | Named people you can talk to directly | Unnamed “team” |
| 10. When would you tell me not to hire you? | An honest answer about fit | “Never” |
What should you ask about agents already in production?
Production proof is the strongest signal you can get. Ask:
- What does the agent do, day to day, for that client?
- How long has it been running?
- How often does a person correct its output?
- What broke after launch, and how was it fixed?
- Can I speak to the client?
A provider who has never shipped a production agent is not automatically wrong for you, but you should know you are their pilot, and the price should reflect it.
What should you ask about the cost to run, not just to build?
Build cost is a one-time number; run cost is forever. Ask for a written estimate of:
- Model usage at your expected volume.
- Tool and platform subscriptions the agent needs.
- Maintenance or support, if any. Published Canadian maintenance arrangements run about $300 to $800 a month.
Then ask what happens to cost if your volume doubles. See the AI agent cost guide for published build ranges.
What should you ask about ownership and access?
- Will the agent run in accounts owned by my business?
- Will I receive the prompts, configuration, and any code written for me?
- What access do you need, and can it be limited to only what the agent uses?
- On the day the project ends, can I revoke your access without breaking the agent?
What should you ask about monitoring and handoff?
- What is included in the first weeks after launch?
- Who on my team needs to be trained, and how?
- What will I see if the agent starts failing?
- What does ongoing support cost, and is it optional?
On personal information specifically, Canada’s privacy regulators expect businesses using AI to have a legal basis for the data, safeguards, and transparency. Your provider’s answers to question 7 should line up with that.
Sources
- Office of the Privacy Commissioner of Canada, AI, privacy, and your business.
- Planwebbies, AI Automation Cost in Canada: Small Business Pricing (2026), published August 5, 2026.
- Innovation, Science and Economic Development Canada, Toolkit for SMEs deploying artificial intelligence.
Next step
Ask EV Advisory the same ten questions. Book a free 20-minute Live Audit, or read what an EV Advisory AI agent implementation project includes first.
FAQ
Questions operators usually ask
What is a red flag in a proposal?
No written success measure, no estimate of monthly run cost, the agent running in the provider’s own account, open-ended hourly billing for the first build, and no mention of documentation or handoff. Any one is worth a question; three or more is a reason to walk away.
Should I own the accounts the agent runs on?
Yes. The tool accounts, model provider accounts, and automation platform accounts should be in your business’s name and on your billing, with the provider given access that you can revoke. That way the agent keeps running if the relationship ends.
How do I check a reference?
Ask for a past client with an agent still running, and call them. Ask what the agent does today, how often a person has to correct it, what broke after launch and how fast it was fixed, and whether they would hire the same provider again.