Hire an AI Developer: What to Check Before You Sign

I get hired as an AI developer, so take this with that bias in mind. But I also inherit a fair number of builds from developers who were hired without these checks, and the same problems show up every time. This is the list I would use if I were on your side of the table.

First, decide what you are actually hiring for

"AI developer" covers very different work. Before you post a job or book a call, write one sentence for each of these:

• The process. What happens today, step by step, and which step you want the model to take over.
• The systems. Which CRM, database, inbox or phone system it has to read from and write to.
• The failure cost. What happens if the model gets it wrong once. An awkward email is cheap. A wrong invoice is not.

If you cannot write those three sentences, any AI developer you hire will spend the first weeks writing them for you, and you will pay for it. If you are unsure what the role covers in the first place, what an AI developer actually does breaks it down.

Eight checks before you hire an AI developer

1. They can show something in production

A demo proves someone can make a model respond. Production proves they can make it keep responding correctly when the input is messy and the API is down. Ask for a live system, a client you can speak to, or public code. Mine are on the systems section of my homepage and the code is on GitHub, and you should expect the same from anyone you shortlist.

2. Their first questions are about your data

Strong AI developers ask where the data lives, how clean it is, who can access it and how often it changes. Weak ones ask which model you want. The model is the easiest decision in the project. IBM makes a similar point about retrieval-augmented generation: the quality of the answers depends on the quality and structure of what is retrieved.

3. They can explain how it will be evaluated

Ask: "How will we know it is working?" A good answer names a test set of real examples, a pass rate, and a review step before every change ships. A bad answer is "we'll test it". Without evaluation, every prompt tweak is a coin flip.

4. They have a plan for when the model is wrong

Every model makes mistakes. IBM's explainer on AI hallucinations covers why. What matters is the design around it: schema validation, confidence thresholds, a human review queue, and logs you can audit. If the developer has not mentioned failure handling by the end of the first call, ask. If they wave it away, end the call.

5. Security is part of the design, not an add-on

Ask how they handle prompt injection, secrets and permissions. The OWASP Top 10 for LLM applications is the checklist I use. Two non-negotiables: the model never holds a credential directly, and every action it takes goes through code that checks permissions.

6. You own everything at the end

The code, the repo, the cloud accounts, the API keys, the prompts and the evaluation set should all be yours. Ask directly. Some agencies keep the build inside their own platform so you cannot leave. I covered that pattern in how to evaluate AI agent development companies.

7. They can estimate running cost, not just build cost

AI systems cost money every time they run. A good AI developer can estimate tokens per run, runs per month and the monthly bill before writing code, then show it on a dashboard after launch. If nobody mentions running cost, it will surprise you later.

8. They tell you where AI is the wrong tool

The most valuable thing an AI developer can say is "this step does not need a model". Rules, lookups and plain code are cheaper, faster and predictable. Someone who wants to put AI in every step is optimising for the invoice.

Freelance AI developer, agency, or in-house?

	Freelance AI developer	Agency	In-house hire
Best for	Defined builds touching a few systems	Large programmes needing many roles at once	AI as a core, ongoing product function
Who writes the code	The person you talk to	Often someone you never meet	Your employee
Speed to start	Days	Weeks	Months
Main risk	Single point of failure if handover is poor	Lock-in and account managers between you and the work	Hard to hire well without someone senior to assess

For most small and mid-sized businesses, a freelance AI developer for the build, with proper handover documentation, is the best value. If the system becomes core to the business, hire in-house later and have the freelancer onboard them. My longer take on the consultant route is in hiring an AI automation consultant.

What moves the price

Rates vary widely by region and seniority, so headline hourly numbers tell you little. What actually moves the cost of an AI build:

• How many systems it touches. Each integration adds authentication, edge cases and testing.
• The state of the data going in. Clean, structured data is cheap. Ten years of free-text notes is not.
• The cost of a mistake. Higher stakes need more validation, review steps and logging.
• Whether a human approves outputs. Approval flows add build time but lower risk a lot.

That is why I do not publish prices and quote only after seeing the stack. Anyone who prices an AI build before seeing your data is guessing, and the guess usually lands on you.

Red flags that should end the call

• They guarantee accuracy figures before seeing your data.
• Everything runs inside their proprietary platform.
• No mention of logging, evaluation or failure handling.
• They cannot show anything running in production.
• They propose a fully autonomous agent for a process that needs one reliable step.

If you want to run these checks on me

I am Ammar Imtiaz, an independent AI developer for hire. I build AI automation, AI agents, CRM integrations and full web applications, and I write the code myself. You can see six systems running in production, read how I work, and book a call to put any of the eight checks above to me directly.

Frequently asked questions

How do I hire an AI developer?

Write down the process, the systems involved and the cost of a mistake first. Then shortlist developers who can show a system running in production, ask about your data before models, explain how the build will be evaluated and how failures are handled, give you full ownership of code and accounts, and estimate monthly running cost.

Should I hire a freelance AI developer or an agency?

For a defined build touching a handful of systems, a freelance AI developer is usually faster and better value, because the person you speak to writes the code. An agency makes sense for large programmes that need several roles at once. Either way, insist on owning the code, accounts and documentation.

How much does it cost to hire an AI developer?

It depends less on hourly rate and more on scope: how many systems the build touches, how clean the data is, how costly a mistake would be and whether humans approve outputs. Ask for a quote after the developer has seen your stack, and ask for an estimate of monthly model and hosting costs too.

What should I ask an AI developer in an interview?

Ask to see something in production, how they would evaluate the system, what happens when the model is wrong, how they handle prompt injection and secrets, who owns the code at the end, what it will cost per month to run, and which parts of your process they think should not use AI at all.

What are red flags when hiring an AI developer?

Guaranteed accuracy before seeing your data, builds locked inside a proprietary platform, no mention of evaluation or logging, nothing running in production to show, and proposals for fully autonomous agents where one reliable automated step would do the job.