- The role: A forward deployed engineer works on site with the customer and takes AI all the way into operation: roll out, connect, adapt, improve. The model comes from Palantir.
- The market: Job postings for “forward-deployed” engineers rose more than 1,000% from January to August 2026 (Lightcast, reported by Fortune); the tech market overall grew 13%.
- The study: McKinsey surveyed 81 executives at software companies. Finding: productivity gains of 30 to 50% in pilots rarely show up in revenue or earnings. FDEs are meant to close that gap.
- The counter-trend: Work at the customer is where AI saves the least. In McKinsey’s data, professional services has the second-lowest AI adoption (26%), because the need for customization is growing.
- For mid-sized companies: Vendors send their FDEs to large accounts. If you are smaller, ask your implementation partner for the same way of working. Five questions for that are below.
What a forward deployed engineer is
The term comes from Palantir. The company sends engineers directly to government agencies and corporations so that its software actually works with the systems already in place. One Palantir job posting compares the responsibility to that of a start-up CTO.
McKinsey defines the role in its study “The AI-centric imperative” of 16 October 2025 as technical experts embedded directly on site with clients to drive successful implementation, integration and scaled adoption of AI. They run pilots, tailor agents to the customer’s workflows and keep optimizing the deployment. According to McKinsey, they shorten the time to value and make it more likely that a solution is used for the long term and expanded.
The difference from familiar roles is responsibility. A consultant recommends and hands over. A sales engineer shows what is possible. An FDE builds it into the real environment and stays until it runs. The measure is not sign-off, but whether people use the system in their daily work.
What the study says, and what it does not
If you quote the study, you should know what it is based on. It is written for software vendors, not for companies that use AI. Its survey figures come from the McKinsey Software CxO AI Impact Survey, Q2 2025, with 81 executives at software companies. Forward deployed engineers take up one paragraph within one of seven shifts, the overhaul of sales and customer engagement. It is not an FDE study. The figures that matter for our topic:
| Finding | Value | Type |
|---|---|---|
| Developer productivity gain in pilots that rarely reaches revenue or earnings | 30–50% | reported |
| Respondents using AI in professional services, or planning to within 12 months | 26% | survey |
| For comparison: the same in R&D | 89% | survey |
| AI platform purchases made in a hybrid model with external partners | nearly 60% | survey |
| Respondents who invested in “technical talent in the field” | 60% | survey |
| Expected cost reduction in professional services | 10–20% | model, illustrative |
Source: McKinsey, “The AI-centric imperative: Navigating the next software frontier”, 16 Oct 2025, text and Exhibits 1, 2 and 5. Survey values refer to software companies.
The rest of the study covers pricing models (away from per-user licences towards billing by consumption or outcome), AI-ready infrastructure and the reshaping of the workforce. For mid-sized companies it is still instructive, because it shows from the other side of the table how vendors intend to sell and roll out AI.
The pilot trap
The most important figure in the study is a caveat. Software companies report 30 to 50% higher developer productivity in pilots. These gains, McKinsey writes, rarely translate into equivalent impact on revenue or earnings. The reason: the tools were dropped into old processes instead of the processes being redesigned.
Anyone who has introduced AI in a mid-sized company knows the pattern. The pilot convinces in the presentation. Then it has to connect to the real system, and it turns out that item numbers exist in two spellings, that approval runs through a person who never appeared in the pilot, and that the team cannot fit the new step into its day. Nobody is responsible, because the pilot team is done and the operations team has not started yet.
This gap is exactly where a forward deployed engineer works. They built the pilot, and they stay for the stretch that follows.
Why work at the customer grows instead of shrinking
One finding runs against the usual story that AI automates everything. In professional services, the work done at the customer, AI adoption is 26%, the second-lowest of ten functions, with only procurement below it. The expected cost reduction of 10 to 20% is the smallest. McKinsey gives the reason explicitly: savings are partly offset by a growing need for customization. In its model of the future workforce, professional services is among the areas that grow again after the cuts.
Behind this is a second thesis of the study. One-size-fits-all applications are being replaced by solutions tailored to industry, customer environment and problem. McKinsey calls this “service-as-software”: platform, agents, automation and expert knowledge as one package that delivers an outcome rather than a tool. The more tailoring, the more work at the customer. And, according to the study, the advantage is shifting from features to data: features can be copied, access to your own good data cannot.
What this means for mid-sized companies
According to Lightcast, a forward deployed engineer in the US earns a median advertised salary of over $188,000. Vendors deploy these people where it pays off: at large accounts. A mid-sized company with 80 licences will not get one. That is not a criticism, it is arithmetic.
According to McKinsey, nearly 60% of AI platform purchases already run in a hybrid model, with external partners for advice and implementation. The study also describes how these partnerships are changing: fewer partners, who get involved earlier, co-develop and stay after the sale to secure adoption. That is the FDE way of working, only outside the vendor.
Our conclusion, explicitly not the study’s: if you want to connect AI to your own data and core processes, ask your implementation partner for exactly this way of working. Not the title, but what it stands for.
How to tell whether your AI partner works this way
Five questions you can ask before you commission anyone. The answers quickly show whether someone goes all the way into operation or stops at the pilot:
- Is it built in your systems or in a demo environment? A pilot on sample data proves little. The problems are in your real master data, permissions and interfaces.
- Who is responsible from requirement to operation? If the team changes between concept and build, knowledge gets lost at that point. Ask for a name, not a department.
- How is success measured? “Signed off” is not a metric. Good answers are: how many cases per week run through the system, how many follow-up questions disappear, how often someone has to step in.
- Who owns the code, configuration and data? If the solution only runs in the partner’s account, you are locked in. Everything should live in your accounts and be documented.
- What happens after go-live? The first weeks in operation show what the pilot missed. An FDE plans for this phase and stays until your team can carry on alone.
When you do not need one
Not every AI rollout needs an engineer on site. If you buy a finished tool that runs without connecting to your own systems, such as a writing, translation or transcription tool, you need a good rollout and clear rules, not integration. This way of working pays off once AI has to access your own data, write into existing systems or prepare decisions in a core process. There, integration decides the value, not the model. How to choose a partner for such projects is covered in our guide to choosing an AI agency.
How we do it
We work this way because we need it ourselves. Our own company runs on the same agents we build for clients, and we know the gap between pilot and operation from our own daily work. What that looks like in practice is described under How we work. We build in your accounts, the code belongs to you, and the models stay swappable.
Sources and method
This article is based on the McKinsey study “The AI-centric imperative: Navigating the next software frontier” by Jeremy Schneider, Joshan Abraham, Matt Linderman and Naveen Sastry (16 October 2025), evaluated from the text and all six exhibits of the PDF version. Survey values come from the McKinsey Software CxO AI Impact Survey, Q2 2025, n = 81, and refer to software companies. Figures on job postings and salaries come from Lightcast data reported by Fortune on 3 September 2026. The transfer to mid-sized companies that use AI, and the five questions, are Digital Maker’s assessment.
FAQ: Forward Deployed Engineering
What is a forward deployed engineer?
A forward deployed engineer (FDE) is an engineer who works on site with the customer and takes software, today mostly an AI system, all the way into day-to-day operation: rolling it out, connecting it to existing systems, adapting it to real workflows and improving it afterwards. McKinsey describes FDEs as technical experts embedded directly on site with clients to drive implementation, integration and scaled adoption of AI. The model comes from Palantir, which sends engineers directly to government agencies and corporations.
How is an FDE different from an AI consultant?
A consultant recommends and hands over; an FDE builds and stays. Consulting usually ends with a concept or a pilot, after which a different team takes over the build. An FDE owns the path from requirement to operation and is measured by whether the solution is used day to day. In many projects, the handover between concept and build is exactly where things stall.
Does the McKinsey study say companies should hire FDEs?
No. The study “The AI-centric imperative” of 16 October 2025 is written for software vendors and is based on a survey of 81 executives at software companies. It describes FDEs as a role vendors use to make their AI products work at the customer. That a mid-sized company, as a user of AI, should ask its implementation partner for the same way of working is our conclusion, not a statement of the study.
How fast is demand for forward deployed engineers growing?
Very fast, from a low base. According to Lightcast data reported by Fortune on 3 September 2026, job postings for “forward-deployed” engineers rose more than 1,000% from January to August 2026 compared with the same period a year earlier, and more than 4,600% compared with 2023. Tech postings overall grew 13% in the same period. The median advertised salary in the US is above $188,000.
Do I need a forward deployed engineer for every AI project?
No. If you buy a finished tool that works without connecting to your own systems, such as a writing or translation tool, you need a good rollout and training, not an engineer on site. This way of working pays off once AI has to access your own data, write into existing systems or prepare decisions in a core process. There, integration decides the value, not the model.
Is your AI project stuck between pilot and operation?
In a discovery call we look at where it is stuck: data, integration, ownership or team adoption. Thirty minutes, concrete next steps, no slides.