Agency, freelancer, in-house or fractional AI lead: who builds your AI
Each option fails in its own way. Agency, freelancer, in-house hire or fractional AI lead, compared on cost, speed, risk and what remains.
5 min read
The choice usually gets made in a hurry. A board meeting asks for an AI plan, someone knows a freelancer, someone else has an agency's deck, and the CTO is asked to hire an AI engineer by next quarter. Six months later, the freelancer has moved on, the hire is still looking for data, or the agency has delivered a demo that works on their laptop and nowhere else.
Each of these can work, and each fails in its own way. The useful question is which failure you can afford.
Short answer: a freelancer for a narrow task, an in-house hire when AI is your product, an agency when you need a product shipped without a team, and a fractional AI lead when you first need someone to decide.
The four options, side by side
Here is how they compare when the goal is a product in production, not a demo.
| Freelancer | In-house hire | Agency | Fractional AI lead | |
|---|---|---|---|---|
| Cost | Lowest day rate, paid while they work | Salary, recruiting fees, months of ramp-up, and someone to manage them | Higher day rate, priced by scope | A few days a month |
| Orders of magnitude | Lowest per day | Highest fixed cost | Varies with scope | Varies with the days per month |
| Time to start | Days to weeks | Several months, between search, notice and onboarding | A few weeks | Days to weeks |
| Main risk | One person: holidays, a better offer, a gap in skills | Hiring the wrong profile before you know the problem | Paying for slides, or for staff rented by the day | No hands to build: it steers, it does not ship |
| What remains after | Code, if someone else can read it | A person who knows your business | Code and documentation, if handover was in the contract | A roadmap and decisions your team can follow |
| Right when | The task is narrow and you can specify it | AI is core to the product for years | You need a product shipped and have no team for it | You need judgement more than headcount |
For what moves the bill beyond the day rate (hosting, cost per request, maintenance), read how much an AI project costs.
In practice, few companies pick just one. These are the combinations we see reach production:
- A lead plus a freelancer. The lead writes the brief and reviews the code, the freelancer builds.
- An agency for the first version, then a hire who takes it over. You recruit for a product that exists, not for a guess.
- An in-house team plus an agency for one deadline. The team keeps the context, the agency brings the hands.
When an AI freelancer is the right call
A freelancer works well when the job fits in one head. Connecting a model to your support inbox, cleaning a dataset, building a prototype to test an idea with five customers.
It works less well when the task touches several systems or several teams. One person cannot be in the product meeting, the security review and the code at the same time, and the gaps show up in production.
Before you sign, ask who reads their code if they leave in three months. If the answer is nobody, keep the scope small. Put the code in your own repository and your own cloud account from day one, not theirs.
When to hire an AI engineer in-house
Hire when AI is part of what you sell, not a tool you use. If your product's value depends on a model you train, tune and monitor every week, you want that knowledge on your payroll.
The trap is hiring first and scoping second. A company recruits a machine learning engineer, the first months go to finding data, and it turns out the first useful project needed a product engineer who knows how to wire an API into the CRM.
Write down the first two projects before you open the role. The job description gets shorter and more accurate. A first AI hire who works alone has nobody to review their work, so plan who checks it before you hire.
When an agency makes sense, and when it does not
An agency is the right answer when you need a product shipped and you do not have the team to build it. A good one has shipped something close to your project before and can show it, works in a written scope, and puts several skills on it at once (product, design, engineering), not one person.
It is the wrong answer in two cases. When the agency sells days rather than a result, you pay for presence, and the bill grows when the project slips. When the people who sold you the project are not the ones who build it, the context you gave in the first meetings gets lost in the handoff.
Three questions sort this out in a first call:
- Who exactly builds it, and will they be on the next call?
- What is in scope, what is out, and when is the first version live?
- What do we own at the end, and can our team run it without you?
Then ask to see a demo in the second week, not a status report.
Our own answers, so you can hold us to them: the people on the first call build it. The scope is written before kick-off, you see a demo every week, and a first version is usually live within 3 to 6 weeks. After launch, we run it with you or hand it over cleanly to your team.
Where a fractional AI lead fits
Many companies are earlier than all of this. They have ten ideas, a budget line, a board that wants progress, and nobody on the leadership team who has shipped an AI product.
That is the gap a fractional AI lead fills. Someone sits in your leadership team a few days a month, picks the projects worth doing, says no to the ones that are not, and checks what vendors and hires deliver.
It also catches the mistakes in the table above before they turn into a hire or a contract. A lead decides whether the next step is a freelancer, a hire or a build, and writes the brief for it.
A good lead's first month ends with a list: every AI idea on the table, the data it needs, who would build it, and a yes or no. Ask for that list. Our fractional AI lead sits in your leadership team a few days a month to keep the roadmap honest.
Start with the question, not the vendor
If you are about to sign with someone, including us, test the brief first. Book a free audit: one hour on your business, then a short document that says what is worth building, and what is not. Then you know whether you need a freelancer, a hire, a lead or a build.
