What an AI agency sells
A team: a project or account manager, designers, engineers, sometimes a data scientist. The engagement usually starts with a paid discovery phase, produces a statement of work, and runs in sprints against it. Changes after that go through change orders.
That structure is the agency's strength when the job is big and you have nobody to do it. It is overhead when the job is one workflow inside a product that already exists. The people who scope the work are often not the people who build it, and the agency's best engineers are spread across several clients.
What a forward deployed engineer sells
One senior engineer, working in your repository, on your data, with your team. The forward deployed engineer scopes the workflow with the person who will use it, builds it inside your product, evaluates it against real examples, and hands it over with what your engineers need to run it: the code, an output rubric, a playbook, and a walkthrough.
The same person scopes and builds, so nothing is lost between the sales call and the code. The limit is capacity: one engineer ships one narrow workflow at a time, not a whole product.
Side by side
| Compared on | AI agency | Forward deployed engineer |
|---|---|---|
| Who does the work | A team, with a management layer between you and the engineers | One senior engineer, working directly with you |
| Where the code lives | Sometimes the agency's stack or hosting, sometimes your repo | Your repository, from the first commit |
| How scope is set | Discovery phase, then a statement of work and change orders | One workflow, written as one sentence, agreed on the first day |
| Time to a first working version | Weeks to months, after discovery | Days, for one narrow workflow on your data |
| Who owns it after | Often the agency, through a retainer, unless the handoff is in the contract | Your team, with a rubric, playbook, and walkthrough |
| Best when | You need a whole product or several features built, and have no team to build them | You have a product and a team, and need one AI workflow live inside it |
When the agency is the better choice
- You have no engineering team. Someone has to run and change the AI feature after it ships. If nobody on staff can, a forward deployed engineer's handoff has no one to hand to, and an agency retainer is the honest answer.
- The job is a product, not a workflow. A new app with design, mobile, backend, and AI all at once needs more hands than one engineer has.
- You need work in parallel. Three features on three deadlines need three builders.
- Procurement needs a firm. Some buyers require a company with insurance, security reviews, and a bench behind it.
When the forward deployed engineer is the better choice
- The product exists and the team is busy. Your engineers can maintain an AI workflow; they don't have the weeks to learn to ship the first one.
- The work is one narrow workflow. One user, one input, one output, and a human check before anything leaves.
- You want to own it. Code in your repo, prompts and evaluation set in your repo, and no retainer to keep it running.
- You want to learn before you hire. One shipped workflow tells you what an AI engineer on staff would actually do, before you write the job description.
Whichever you pick, put this in the contract
- The scope as one sentence: who uses it, what goes in, what comes out.
- Acceptance criteria, written before the build, that say what a correct output looks like.
- An evaluation set of real examples, so a prompt change can be tested instead of argued about.
- A handoff your team can run without the vendor: documentation, a rubric, and a recorded walkthrough.
- A data processing agreement with your model provider, such as Anthropic or OpenAI.
- Plain ownership: the code, prompts, and evaluation set are yours.
The FDE Sprint is the fixed-scope version of the forward deployed route: one workflow, a working prototype on your data by Day 5, in your repo by Day 10, at a fixed price with a full refund if either milestone is missed. To compare it with hiring and contracting on cost, use the FDE cost calculator.
Frequently asked questions
Should I hire an AI agency or a forward deployed engineer?
Hire an agency when you need a team to build something you don't have an engineering team to own: a whole product, or design, mobile, and backend at once. Bring in a forward deployed engineer when you have a product and a team, and you need one AI workflow shipped inside your codebase that your engineers will run afterward.
Is a forward deployed engineer cheaper than an AI agency?
For a single, narrow workflow, usually, because there is one senior engineer instead of a team with a project management layer, and no long discovery phase. For a large build, compare the quotes on cost per shipped, accepted workflow rather than on day rates, and count the handoff: work your team can't maintain is not finished.
What should an AI development contract include?
One sentence of scope agreed before work starts, acceptance criteria that say what a correct output looks like, an evaluation set to test against, a handoff your team can run without the vendor, a data processing agreement with the model provider, and a clear statement that the code and prompts belong to you.
Not sure your workflow is narrow enough? Score it in two minutes, then book a 15-minute Scoping Call.