Hiring for AI: freelancer, agency, or in-house?

Freelancers are cheapest and riskiest on handover. Agencies cost more and often hand you juniors. In-house is right only once you know exactly what you're building. Most pre-product-market-fit teams should buy a fixed-scope build, not a person.

There are three ways to get an AI product built, and the honest answer about which is right depends almost entirely on how well you already understand the problem.

Freelancer

Costs: $25 to $120 an hour depending on where they are and how senior.

Works when: the scope is genuinely clear, you can review the code yourself, and the project is small enough that one person can hold all of it.

Fails when: they disappear. Not maliciously, a better contract comes along. What you're left with is the part nobody thinks about at the start: no documentation, a deploy process that lived in one person's terminal history, and a system nobody else can run.

If you go this route, make handover a deliverable with its own line item, not a favour you ask for at the end.

Agency

Costs: typically $15,000 to $150,000 per project.

Works when: you need several disciplines at once and have the budget for coordination overhead.

Fails when: you meet a senior person in the sales call and get juniors on the build. This is common enough that you should ask directly: who specifically writes the code, and can I talk to them before signing? A straight answer takes two seconds. A vague one tells you what you need to know.

Also ask who owns the code. If the word "license" appears, or there's a pause, you're renting a dependency rather than buying an asset.

In-house

Costs: $120,000 to $200,000 a year, plus three months to hire.

Works when: AI is core to the product long-term and you already know what you're building.

Fails when: you hire before the problem is defined. A strong engineer with an unclear brief produces an expensive prototype and leaves within a year, because good people leave roles where the goalposts move.

The question that actually decides it

Not budget. This one: can you write down what "working" means, specifically enough that someone could test it?

If yes, buy a fixed-scope build. You know what you want, so pay a fixed price and take delivery.

If no, don't hire anyone yet. Spend two weeks getting clear. Whatever you buy while the problem is fuzzy, you will pay for twice.

What a good arrangement looks like

• A fixed scope and a fixed price, agreed before work starts
• Direct access to the person building it
• 100% of the code and IP transferred on delivery
• Documentation and a runbook someone else can deploy from
• An honest answer when the thing you asked for is the wrong thing

That last one is worth more than it sounds. The most valuable thing a builder can tell you is that you shouldn't build it yet.