How to choose an AI agency
A good AI agency leaves you with a working system that gives your team hours back every week. Plenty of projects end as a demo nobody uses, or a chain of no-code steps that breaks when a tool changes. These questions help you tell the difference before you sign.
Deniz SancarFounder of Yarigai. Updated 10 October 2026.
Yarigai is an AI agency, so read this with that in mind. The advice works for any agency you talk to, including us.
Start with the job, not the AI
Before you talk to any agency, write down the one job that costs your team the most hours each week. Count the people who do it and the hours each of them spends. That number tells you what a fix is worth, and it gives the agency something concrete to quote against.
A good agency will ask you for it on the first call. If an agency pitches an AI strategy before it asks what your team does all week, keep looking.
The three kinds of AI agency
Strategy consultancies
They write the plan, usually through workshops, a roadmap and a list of use cases. That helps if your board needs a plan. You still need someone to build it.
Automation shops
They connect your tools with no-code platforms like Zapier, Make or n8n. That's quick to start and fine for simple steps. As the workflow grows it gets harder to test and maintain, and you pay the platform per task.
Build agencies
They write real software, with code, tests, hosting and documentation. It takes longer to start than a simple zap, and the result runs on infrastructure you own. Yarigai is a build agency.
Ten questions to ask before you sign
Who owns the code, the data and the accounts?
You should, from day one, with everything hosted in your name. Be careful if the system lives on the agency's own platform and stops when you leave.
What will it cost to run each month?
Every AI system has bills for AI usage and hosting. A good agency estimates them before you commit and designs to keep them low, for example by sending routine steps to smaller models.
Can I see something you built that's live today?
Ask for a link you can use yourself, not a slide or a recorded demo.
What happens when the AI gets it wrong?
Look for human approval on anything that sends money, emails a customer or deletes data, plus a log of every action the system takes.
Will I see working software during the build?
A weekly demo of the real thing catches problems early. A single reveal at the end hides them.
Is the price fixed?
A fixed quote after a short discovery phase puts the risk of overruns on the agency. Open-ended hourly billing puts it on you.
Will it work inside the tools we already use?
Your team shouldn't need to learn a new app to get the benefit. The systems that get used work inside the email, CRM and chat tools people already open every day.
How will you test it before launch?
A good answer involves your real past data, like running the system on last month's emails and comparing its work with what your team did.
How will you handle our data?
Ask which AI providers they use, whether those providers can train on your content, what access each system gets, and whether they'll sign an NDA and a data processing agreement.
What happens after launch?
You should get documentation good enough for another developer to take over, plus an optional support plan you can cancel.
Red flags
- The agency can't show you anything live.
- The proposal charges a per-seat licence for a thin layer on top of ChatGPT.
- Nobody can tell you what the system will cost to run each month.
- Every problem gets the same answer, usually a chatbot.
- You don't own the code or the accounts.
- The project ends with a strategy document and no working software.
How long a good AI project takes
A common shape is a free scoping call, then about a week of discovery that ends in a written plan and a fixed quote, then two to six weeks of build with weekly demos. A single workflow can go live in about two weeks.
If an agency needs three months before you see anything working, ask why.
How Yarigai answers these questions
- You own the code, the data and the accounts from day one.
- The monthly running cost is on the quote, and you pay AI and hosting bills at cost.
- A person approves anything that sends money, emails a customer or deletes data.
- You see working software every week, tested on your own work.
- You get a fixed price in writing after a one-week blueprint.
- We built and still run 38-0-0, Mockingbird, Outtest, Fighter Dynasty and Basketball Dynasty. You can use every one of them today.
Common questions
What does an AI agency do?
An AI agency plans, builds and often runs AI systems for a business. Depending on the agency, that means a strategy document, no-code automations or custom software such as AI agents, internal tools and support bots built into the tools the business already uses.
How long does it take an AI agency to build a system?
A typical project has about a week of discovery, then two to six weeks of build. A single workflow can go live in about two weeks.
Should I hire an AI agency or an AI consultant?
Hire a consultant if you need a plan and have people to build it. Hire a build agency if you want the system built and running. Some agencies do both, so ask whether the deliverable is a document or working software.
Who should own the AI system an agency builds?
You should. Ask for the code, the data, the prompts and the accounts to be in your name from the start, so the system keeps running if you change agency.
Ask us the ten questions.
Book a free scope call and hold us to every answer.