Freelancer, agency or forward deployed engineer: who should build your AI project?
An honest comparison of the three ways to get an AI project built: what each costs, how each works with your team, and which one fits which kind of project.
Hussain CFreelance AI engineer in Surat, IndiaThere are three common ways to get an AI project built if you don't have the team in-house. You can hire a freelancer, hire an agency, or bring in a forward deployed engineer who works inside your team for a while.
I've been all three at different points, and I've competed with all three. Each is the right answer for some projects and the wrong one for others.
How do they compare?
| Question | Freelancer | Agency | Forward deployed engineer |
|---|---|---|---|
| Works from | A spec you write | A spec they write with you | Inside your team, in your Slack |
| Best for | Well-defined builds | Large, multi-discipline projects | Fuzzy problems inside a real workflow |
| Cost | Lowest | Highest | Similar to a senior freelancer |
| Speed to start | Days | Weeks | Days |
| Who you talk to | The person doing the work | An account or project manager | The person doing the work |
| After launch | Varies a lot | Support contract | Stays through the messy part |
| Main risk | One person, one point of failure | Juniors on your project | Needs your team's time |
When should you hire a freelancer?
When you know exactly what you want. "Add a summarise button to our ticket view, using our existing API, by the end of the month" is a great freelance project. You write the spec, a good freelancer builds it, and you review the result.
It goes wrong when the spec is really a question. "Make our support smarter with AI" handed to a freelancer usually ends with a demo that answers questions nobody asked.
When should you hire an agency?
When the project is big and needs more than engineering: design, research, several developers, QA and someone to coordinate them all. A new product with a mobile app, a web dashboard and an AI feature is agency territory.
Ask who will actually work on it. The senior people you meet in the sales call may not be the ones writing your code. That isn't always bad, but you should know.
When is a forward deployed engineer the right fit?
When nobody can write the spec yet, because the problem lives inside how your team works. AI projects are often like this. The value is in knowing which part of the workflow to hand to a model, where the data actually lives, and what your staff will and won't trust.
- 01Sit with the teamWatch the work as it's done today
- 02Scope one thingThe smallest piece worth real money
- 03Build itIn your repo, with your engineers
- 04Run it liveFix what real use uncovers
- 05Hand overDocs, training, code your team owns
The catch is that it needs your time. Someone on your side has to answer questions, open doors and make decisions. If nobody can, a forward deployed engineer turns into an expensive freelancer.
Which one fits your project?
- Can you write a one-page spec that a stranger could build from?Yes →Hire a freelancer.No ↓
- Does it need design, several engineers and project management at once?Yes →Talk to agencies.No ↓
- Is the hard part understanding your workflow and data?Yes →Bring in a forward deployed engineer.
- Still no →Start with a short paid scoping exercise with any of the three, and decide after.
Whoever you choose
A few rules apply to all three:
- Code in your repository, from the first commit. Not on their laptop, not in their account.
- A written scope, with what's out as well as what's in.
- A demo on your real data within the first two or three weeks.
- A named person who's accountable after launch.
I work as a forward deployed engineer, so take my view with that in mind. If your project is a better fit for a freelancer or an agency, I'll tell you. Here's how to reach me.