Alternatives to hiring an in-house AI team
Four routes to the same capability, ranked by what each one leaves you owning when it ends.
The verdict
Build an in-house AI team when AI is your product and you already run engineering well, use an agency when the first system has to reach production before you know what to hire for, and buy off the shelf when the problem is genuinely the same as everyone else's.
The case against hiring first is sequencing, not capability. Writing a job description for an AI engineer requires knowing what your AI system will need, and most teams learn that during the first build.
There is also a quieter risk. One AI hire holds every decision, prompt and workaround in their head until a second one arrives. That is a fragile arrangement to depend on for something customer-facing.
- The buildAny route can produce this
- The judgementWhich approach, and what to skip
- The eval suiteProof it still works next month
- The operatorSomeone paged when it breaks
- The contextWhy each decision was made
Hiring buys all five, eventually. The routes below buy them in a different order.
How we judged these
Verified
We are an agency, so one row here is us. The criteria are applied the same way to every route, and the ranking is by fit rather than by preference.
Third-party figures come from MIT's 2025 State of AI in Business report, published July 2025. Everything about our own work comes from systems we run.
- Time to a working system
- How long before something real is in front of users.
- Where the context ends up
- Who understands why the system works the way it does, a year later.
- Failure mode
- What specifically goes wrong with this route, and how visible it is.
- Exit cost
- What you keep if the arrangement ends.
The field
| Route | Starts | Context lives with | Typical failure |
|---|---|---|---|
| In-house team (incumbent) | After a hiring cycle | Your employees | The first hire leaves before the second arrives |
| AI agency | In days | The agency, unless you require handover | A build handed over with nothing written down |
| Embedded engineers | In weeks | Shared, if your team is in the work | Treated as extra hands, so nothing transfers |
| Marketplace freelancer | In days | One individual | The engagement ends and takes the reasoning |
| Buy a product | Immediately | The vendor | It does 80% and the last 20% is your business |
Ranked, by what you need first
- 01
An AI agency, with handover written in
First system built, knowledge transferred
The right first move when you have a problem worth solving and no way to specify the role yet. You get a working system, and the build teaches you what the eventual hire needs to be good at.
MIT's 2025 State of AI in Business report found pilots built with an outside partner reached full deployment twice as often as internal builds. Employee usage was nearly double. The condition is the handover. Require the evaluation set, the deployment steps and the reasoning as deliverables, or you have rented capability rather than bought it.
Best for
- A first production AI system, where the requirements are still moving
- Regulated work needing an audit trail from day one
- Teams who want the option to bring it in-house later
Not for
- Work that changes daily and needs someone in every standup
- Organisations that will not commit to owning it eventually
- 02
Embedded engineers alongside your team
Capability transferred while building
Choose this when you have engineers who are strong but new to AI work. They keep ownership, and the specialist supplies the parts that are unfamiliar: evaluation design, retrieval quality, failure handling.
The failure mode is treating it as extra hands. If your engineers are not in the work, nothing transfers and you have paid a premium for contractors.
Best for
- A capable engineering team missing only AI experience
- Organisations that intend to own the system permanently
Not for
- Teams with no spare engineering capacity to pair
- Buyers who want an outcome rather than a collaboration
- 03
Buy an existing product
No build at all, where it fits
Always worth checking first. Where your problem is the same as everyone else's, a product beats anything custom on cost and time. The vendor absorbs the maintenance too.
The catch is the last stretch. A product covers the common path, and the exceptions specific to your business are usually where the value was. Buy when the exceptions are rare, and build when they are the point.
Best for
- Standard problems with mature vendors
- Teams who want a result without owning any of it
Not for
- Workflows that are the reason customers choose you
- Anywhere your data cannot leave your own systems
- 04
One senior freelancer
Fast start, thin foundation
Reasonable for a contained piece of work with a clear finish. It is fastest to start and has the shortest reach: everything the system knows about itself ends up with one contractor.
It gets chosen for the wrong reason often. When the real need is a production system somebody will still be accountable for, this route ends at exactly the wrong moment.
Best for
- A prototype meant to answer one question and then stop
- Filling a narrow gap on work your team already runs
Not for
- Anything customer-facing you cannot afford to have quietly degrade
- Teams with nobody able to review the work
First AI systems taken to production
Common questions
01What does moving from an agency to an in-house team actually take?
It takes a written handover and an overlap period. The code is the easy part. Four things have to transfer: the evaluation set, the deployment path, the failure history and the reasoning behind past decisions. Ask for those as deliverables during the build, not at the end.
02Is one AI engineer enough to start?
One engineer can build a system and cannot safely operate it alone. Holidays, illness and resignation each remove the only person who understands it. If the budget stretches to one, spend part of it on writing things down.
03How do we know when to stop outsourcing?
Stop when your own team can explain why the system works the way it does. That is a sharper test than headcount or spend, and it is usually reached during the second or third significant change rather than at launch.
04Should the first AI hire be an engineer or a lead?
Hire the engineer first in almost every case. A lead with nobody to lead spends a year writing strategy documents. What teaches you who to hire next is a system in production. Bring in the lead once there are two or three engineers to coordinate.
Related
- agency against an in-house AI team →The direct comparison of two of these routes.
- staff augmentation or outsourcing →The embedded route, in detail.
- hire a dedicated development team →What the embedded arrangement looks like here.
- what happens after the AI pilot →The moment this decision usually arrives.
