- An FDE works where the work happens, writes and fixes the system based on what they see, and is done when your team can use the system without them.
- A first round usually takes about a month: week one to listen and measure, week two for something to try, the last two weeks to refine and hand over.
- It pays off when the company has AI tools but few people really use them, and nobody knows yet what to build for whom. If you already know and can write a clear spec, a project-based build fits better.
- An executive who hires an FDE has to let frontline staff spend time talking and testing every week. If you can't give that time, don't hire one yet.
What an FDE is, seen from the company paying for one
From the buyer's side, an FDE differs from other outside teams in three ways. They sit at the same desks as the department that will use the system. They write and fix the system themselves every day, based on what they see on the floor. And their work ends when your team can run the system alone; a summary report is a by-product.
The term comes from Palantir, which placed engineers at client organizations from its early years. More recently, OpenAI and Anthropic have also hired FDEs to help enterprise customers put their models to work. The reason is the same: however good the software, it sits on the shelf unless someone adapts it to how the client works.
If you searched for the term and found only job ads, that's because most of what's written about it is for people applying for the role. The rest of this page answers the hiring side: what to prepare, and what you should get back each week.
What an FDE does each week
An FDE's first round usually runs about four weeks. For each week below: what they do, what your side needs to provide, and what should be on the table by Friday.
Week 1: listen, watch and measure
The FDE sits with two or three teams the executive chose, watches the real work, asks which steps waste time, and times the current way of working for comparison. Your side provides one executive who can make decisions, frontline staff with time to talk, and access to the data needed. By Friday you should have a ranked list of tasks and baseline numbers for the one chosen.
Week 2: a first version people can try
The FDE builds a first version for one task, small enough for two or three users to try on real work within days, then adjusts it almost daily based on what they see. Your side has users try it for real and say plainly what's awkward. By Friday you should see people on the team using the system for that day's work.
Week 3: handling the exceptions
This is when the work that doesn't fit the pattern shows up: unreadable documents, customers with special terms, data in the wrong place. The FDE connects the system to the missing data, sets permissions to match job roles, and starts training the lead users who will teach their colleagues. Your side provides someone who can say how each special case should be handled.
Week 4: measure and hand over
Compare the numbers with week one, write the manual, hand all code and accounts to the system owner on the company side, then decide together whether to move to the next task, extend this one to another team, or stop. By Friday there should be a system the team uses, before-and-after numbers, and someone in the company who can look after it.
After the first round, a good FDE steps back gradually, from sitting with the team daily to dropping in weekly, then to answering questions only. If months later the team still has to call them every time the system hiccups, the handover isn't finished.
For the development team · Access and tools an FDE should get
Give the FDE a named account in the company's SSO, tied to permission groups. Start with read-only access to the data the first task uses, test on a staging environment separate from production, keep the code in the organization's repository from day one, and keep passwords and API keys in the company's secrets store. On the last day of the round, disable the account and rotate any keys the FDE has seen.
How an FDE differs from a consultant, an AI engineer and hiring your own team
All four routes can get a company using AI. They differ in who decides what to build, who writes the system, and where the knowledge sits when the work ends.
| Topic | FDE | Consultant | AI engineer (contract build) | Your own team |
|---|---|---|---|---|
| Where they work | In your company, next to the users | Meeting rooms and workshops | At the contractor's office, from the spec you send | In your company |
| Who decides what to build | Found together with users along the way | The consultant proposes, executives choose | You write a clear spec before work starts | Your team lead |
| Do they write the system? | Yes, and adjust it on site almost daily | Mostly no | Yes, but at a distance from users | Yes |
| What you get | A system the team already uses, with a manual and an owner | A plan and recommendations | A system built to spec | Depends on who you manage to hire |
| When results show | Within weeks | After the plan is done and someone builds it | When the spec is delivered | After hiring and learning your systems, often months |
| Where the knowledge sits afterwards | With lead users and the system owner on your side | In the report | With the contractor, unless the contract says otherwise | In the company, as long as people stay |
| Fits when | You want to use AI but don't yet know what to build for whom | You need a company-level direction first | You know exactly what system you need | AI is long-term core work with a steady pipeline |
| Main risk | Relying on outsiders too long without a handover plan | A good plan nobody builds | Built to spec, but nobody uses it | Hard to hire, and slow |
Many companies use more than one route: a consultant to set direction, an FDE for the first two or three tasks, then their own team once they know what kind of people they need. For the big picture of where company-wide change starts, read What is AI transformation? Where to start alongside this.
When an FDE pays off, and when it doesn't
An FDE puts a senior engineer's full time into your team. What you get back is the time you don't waste building things nobody uses. If your company doesn't have that problem, another route may fit better.
Signs an FDE is likely worth it
- You've bought AI licenses, but few people really use them and nobody knows where they get stuck
- The work you want AI to help with is spread across several departments, each doing it differently
- IT is busy with existing systems and has no time to sit with users
- Executives are ready to let frontline staff spend several hours a week talking and testing
- You want people inside the company to use and maintain the system after the round ends
There are five cases where you shouldn't start with an FDE.
- You know what system you need and can write a clear spec. A project-based build fits better, and its scope is easier to control.
- No executive has agreed to own this. The FDE will circle around questions nobody answers.
- Frontline staff have no time at all, and an FDE works by sitting next to users.
- You need someone running AI systems full time for years. Building your own team pays off better in the long run.
- Most of the data the task needs is still on paper or in people's heads. Fix the data first.
Hypothetical case
A freight company with every license, and Monday mornings unchanged
Sample data (names and figures invented): a mid-sized freight company bought Copilot for 80 office staff. Six months later, about a quarter of them open it every week. Two training courses have been run. The executives decide to try one FDE for a month.
In week one the FDE sits with the dispatch coordinators and customer service, and finds that every morning the coordinators spend about two hours combining photos of delivery notes sent by drivers with Excel files from the warehouse, to build summaries for major customers. This work never came up in the training, because nobody thought of it as something AI could help with.
Week two has a first version that reads the delivery-note photos and fills in a summary table for a person to check. Week three adds a list of unreadable photos for a person to review first, and trains two shift leaders as lead users. By week four, the morning summary takes about 40 minutes, the manual and code are with the company's IT team, and the next round targets customer service's parcel-tracking emails.
The Friday note: one page that keeps an FDE accountable
Agree on day one that every Friday the FDE sends a one-page note to the executive who owns the work, under four headings.
- Who used the system for what this week
- The chosen task's numbers, compared with week one
- What's stuck waiting on the company side, such as access, data or an answer about a special case
- What happens next week, and what's been dropped
If item 1 is empty two weeks running, stop and talk before going further.
Questions executives ask before hiring an FDE
How is an FDE different from a solutions engineer or sales engineer?
Solutions engineers and sales engineers work before the sale, demonstrating the product and answering technical questions to help the customer decide. An FDE arrives after the contract is signed, works inside the customer's team, and stays until the system is in real use.
How much of our data does an FDE need access to?
Only what that round's work needs. Start read-only, through a named account the company can disable when the work ends, with a confidentiality agreement and a data processing agreement under PDPA in place before day one. An FDE should work under the company's AI usage policy like any employee. If you don't have one yet, see the topics in An AI usage policy your staff will follow.
Who looks after the system when the round ends?
The system owner and lead users on the company side, using the manual the FDE wrote. Agree before you start who those people are and how long a question-answering period follows the round. If the company needs someone full time, that's a signal to start building your own team.
Can an FDE work remotely?
The building can happen anywhere, but week one and the handover should be on site, because most of what an FDE needs to see isn't in any document: the file staff keep open all day, or the person everyone walks over to when something odd comes up.
Can a mid-sized company use an FDE?
Yes, if there's a clear owner and two or three tasks that repeat every week. Mid-sized companies often start with one FDE and one team in the first round, and tend to decide faster than large enterprises.
If you want an FDE sitting with your team
Your frontline staff know how the work really runs, which cases are exceptions and which customers need special care. Our FDEs start by listening to those people, then turn what they hear into process steps, rules and decision points a system can follow. We don't decide for your people what's right in accounting, sales or production. We turn their decisions into a system that runs the same way every day.
DNA Maker's FDE team service places senior engineers from our Bangkok team inside your company in rounds. We design and write the systems ourselves, with no subcontractors, and track the same numbers with you from the first week. If you're not sure whether an FDE or a project-based build fits better, send us two or three tasks you'd like AI to help with, and we'll tell you plainly which is the better deal for your company.
SOFTWARE ENGINEERING GLOSSARY
Software engineering glossary
| Term | What it is | A simple example | What executives should ask the development team |
|---|---|---|---|
| Forward Deployed Engineer (FDE) | An engineer a vendor places inside a client company to build and deliver systems from where the work happens | An engineer at the same desks as the dispatch team for four weeks | Is the person sitting with our team the same person who writes the system? |
| Discovery | The first phase of the work, spent understanding the problem before deciding what to build | Week one, when the FDE watches the work and times it | What document does discovery end with, and what do we approve before building starts? |
| Champion | A lead user on the team, trained to teach colleagues and answer their questions in place of the FDE | The shift leader everyone walks over to when the system stalls | Once the FDE leaves, how many hours a week does the champion have for this? |
| Read-only Access | Permission to view data but not change or delete it | The FDE can view orders in the ERP but can't cancel one | If more access is needed later, who approves it? |
| Repository | Where all of a system's code and its change history are stored | Your organization's GitHub or GitLab account | Is the code in our company's repository from day one? |
| Staging Environment | A test copy separate from the live system, for trying changes before users see them | Having the AI summarize yesterday's data without touching the system in use | What has to pass through staging before it goes live? |
| Handover | Passing the system, documents, access and knowledge to the people who will look after it | On the last day of the round, the FDE disables their own account and gives the manual to the system owner | If the FDE didn't show up tomorrow, what would we already have in hand? |
