Why forward deployed engineering is growing

The role existed with little attention for about twenty years. Then, across roughly eighteen months, it became one of the most-discussed jobs in the industry. Three things caused that, and the numbers behind each are worth handling carefully, because this is a topic where figures get repeated without their sources.

Cause one: pilots that do not reach production

In July 2025, MIT's Project NANDA published "The GenAI Divide: State of AI in Business 2025". Its headline finding was that 95 percent of generative AI pilots delivered no measurable profit-and-loss impact, and that only 5 percent of integrated systems created significant value.

That number is now quoted everywhere, usually without its sample size. Here is the caveat it deserves: the study drew on 52 executive interviews, surveys of 153 leaders, and analysis of 300 public deployments. That is a real piece of research, and it does not cover every company. Treat 95 percent as a strong signal about a real problem rather than a precise measurement of an industry.

What makes it credible is that it matches what anyone selling AI to a large company describes. The model works. The pilot impresses people. Then it stops before it becomes part of the actual daily work, and the reasons are almost never about the model. They are about data nobody had audited, a success metric nobody defined before the work started, and users who were not involved in designing the workflow they were handed.

None of those three are solvable from outside the building. That is the whole argument for the role. Forward deployed engineering for enterprise AI covers that final step before production in detail.

Cause two: investors started arguing for it in public

On 4 June 2025, a16z partner Joe Schmidt published "Trading Margin for Moat", which called the forward deployed engineer the most in-demand job in startups and argued that AI companies should build service teams focused on implementation instead of trying to grow through the product alone.

The evidence he uses is the previous generation of enterprise software. ServiceNow's gross margin at IPO was 63.2 percent and Workday's was 54.1 percent, both well below the 80 percent that software investors treat as normal. Both later rose, to 79 percent and 75 percent in 2024. The early focus on services was the way in, and the software business came after it.

The piece also noted that OpenAI had 22 open forward deployed and solutions engineering roles out of 311 total positions at the time of writing, which is a useful sign of how seriously one of the leading AI labs was treating this way of selling in mid-2025.

When a firm of that size publishes an argument, founders act on it. A good part of the growth in postings is companies following advice they read.

Cause three: the largest cloud provider made it a product

In June 2026, AWS announced a Forward Deployed Engineering organization backed by a $1 billion investment, embedding thousands of engineers with customers to build production AI systems.

Two details in that announcement matter more than the dollar figure. AWS said the work would be structured around shared goals and business results rather than billable hours. And it named customer self-sufficiency after the engagement as a design goal. Francessca Vasquez, VP of Frontier AI Engineering and Services, framed the demand as customers having moved past exploring what AI can do and wanting it to be core to how they operate.

A month later, AWS extended the model to consulting partners, describing a three-phase pattern: AWS engineers embedded with partner teams, then support shifting toward helping partners scale their own practice, then partner teams working independently. It also named the reusable assets that make the model repeatable, including domain ontologies, evaluation frameworks and capability registries.

When the largest existing cloud provider builds an organization around embedded engineers and prices it on outcomes, the model is no longer an unusual practice of one company.

What the hiring data shows

The clearest measurement comes from the analysis of 1,000 postings published in November 2025 using Revealera data. Postings carrying the title grew 1,165 percent year over year, comparing January to October 2025 against the same period in 2024, and October 2025 was the highest volume on record.

Two things in that dataset qualify the growth. First, the title covers three different jobs, so part of the increase is relabelling rather than new work. We explain that in detail in the three different jobs that use the title. Second, 58 percent of postings came from companies with 11 to 200 employees, which tells you the adopters are growth-stage companies selling to large enterprises rather than only the well-known names.

The industries named are consistent with the argument: financial services in 24 percent of postings that specified one, government and defence 18 percent, healthcare 17 percent, insurance 17 percent, energy and utilities 13 percent. These are the sectors where the environment, rather than the model, is the hard part.

The single cause behind all three

Leave out the announcements and one cause remains. Every large company has its own data, its own workflows, its own compliance constraints and its own definition of good enough. Selling an AI product to one of them is also selling an integration project. Somebody has to do that project.

For twenty years that somebody was a systems integrator working from a specification. What changed is that the specification cannot be written in advance any more, because with AI systems the customer does not know what good looks like until they see the system behave on their own data.

What could stop it

Worth saying, since most coverage of this treats the trend as permanent.

Two things would slow it. The first is the model's own economics: an embedded engineer is expensive and reduces gross margin, which investors see, and a16z's Marc Andrusko argued in January 2026 that most companies copying this end up with an expensive services business rather than a platform. The Palantirization problem covers that case.

The second is automation of the work itself. Palantir now sells a product called AI FDE, an agent that operates its platform through conversation. The part of the job that is platform operation is being automated by the same technology that created the demand. Is the role being automated explains what work remains.

Our view: writing the code takes less time, and the accountability stays with a person. Deciding which problem is worth solving inside an organization that cannot describe it stays a job for a person, and so does being answerable when the deployment does not run.