Forward deployed engineer: hire one or use an agency?
Hiring a forward deployed engineer makes sense when the need is lasting, central to the business and large enough to keep one person busy full time. An agency makes sense when you want a first result fast, on a bounded need, without building a team. Either way, ask the same questions: who builds, who owns the code and the data, what security applies, what the first result is, who follows up, and how you exit.
Two routes lead to a forward deployed engineer, an engineer who works close to your company, on your data and your processes: you hire one, or you go through an agency that puts one at your disposal. Each wins in some cases. This article gives the checklist to run through before choosing, then the situations where hiring in-house is the better decision.
Who actually builds it?
Ask for the names and backgrounds of the people who will write the code, not only the person who signs the quote. In an agency, the person who sells and the people who deliver are not always the same. Ask how many projects each engineer runs at once, and whether they will be on your premises or on video calls.
On the hiring side the same question runs the other way: one person hired is one point of failure. If they leave or fall ill, the project stops.
Who owns the code, the models and the data?
These are three separate subjects. For code and models, ask in writing who owns them at the end, whether you can access them during the project, and whether they can run without the agency. For data, ask where it is copied, how long it is kept, and what happens when the contract ends.
An employee settles most of these points through an employment contract. With a supplier they are settled in the service contract, and a lawyer should read that part.
What security should you ask for?
The basic questions are the same for any supplier: where the processing is hosted, who can access it, whether the AI model trains on your documents, whether everything can be deleted on request, and whether the solution can be installed at your site. Ask too whether the supplier only reads your sources or copies them. Vague answers, or answers postponed to later, are a signal.
What is the first result, and when?
A good project quickly shows something running on your real data, in the hands of the people who will use it. Ask for a date and a precise description of what you will see on that date. Be wary of a first deliverable that amounts to an audit or a presentation, unless that is what you asked for.
Ask also how the business side validates. Nothing should go to a customer or a partner without the approval of someone at your company.
What happens after go-live?
A system in production needs adjustments: a document format changes, a tool is replaced, the team wants to refine an instruction. Ask who answers when something breaks, through which channel, and whether follow-up is included in the price or billed separately. A named contact beats a ticket desk.
How do you leave?
Ask before signing. What is the commitment period, what is the notice, what do you take away when you go (code, documentation, access, data), and in what form. An agency confident in the value of its work has no need to lock you in.
When is it better to hire in-house?
| Situation | Why in-house wins |
|---|---|
| The need is permanent and central to the business | Knowledge builds up inside the company |
| There is work for one person full time for a long period | An employee is often the better value over the long run |
| No third party can see the data | Outsourcing is not an option |
| You already have a technical lead to manage the role | They can set priorities and review the work |
In these cases hiring is the right call, even if recruiting takes time. A common compromise is to start the project with an agency while you recruit, provided the contract covers handing the work over to the person you hire.
If you want to test a use case, if the need is bounded, or if nobody inside yet knows how to manage this kind of profile, an agency spares you from building a team before you know what you need.
And with us?
We are a forward deployed agency, so we are a party to this question: read the checklist above as applying to us too. We are three Télécom Paris engineers specialised in AI. We build custom agents and also take machine learning and data missions, for the automotive sector (a network of garages), environmental engineering and consulting.
In practice: a first 30-minute call, free and without commitment. Access to your tools is authorised in two clicks, with no password, read-only on SharePoint, OneDrive or Dropbox in the folders you choose, with no file kept. Two weeks after the call, a first agent is running, and a dedicated contact follows up afterwards. Nothing goes out without the validation of someone at your company.
On security: one encryption key per company, servers in France, a database in the European Union, European models (Mistral AI) under a no-training contract, installation at your site possible, and erasure on request. The commercial model is a set-up fee based on the volume of documents, then a monthly subscription for the whole company with unlimited users; amounts are given on the call.
To compare formats, see the page on agency, freelance or subscription and the one on the AI integrator for companies.
Questions
When is it better to hire a forward deployed engineer in-house?
When the need is lasting and touches the core of the business, when there is enough work to keep one person busy full time for a long while, or when the data is so sensitive that no third party can access it. You also need someone able to manage them technically.
What should you ask an agency before signing?
Who concretely builds, who owns the code, the models and the data, where the processing runs, what first result you will see and when, who follows up afterwards, and how you leave the contract with your work in hand.
How do you judge a first result?
It should run on your real data, be tested by the people who will use it, and be shown before any slide deck. A deliverable made only of slides proves nothing.
Can an agency engagement be reversible?
Yes, if you ask before signing: access to the code, documentation, data that stays with you, and a clear exit clause. An agency that refuses to answer these points has already told you something.