Why the FDE Market Exploded | 2026 in the Numbers on Jobs, Investment, Acquisitions, and Results

Hello, this is Ryuta Hamamoto from TIMEWELL.

This is the eighth article in our FDE practice series. The first seven covered the skills used on site and how a week is run. This one changes the angle and reads, in numbers, why this role expanded so suddenly from 2025 to 2026. The article What is an FDE listed the milestones; here I narrow it to four sets of numbers, jobs, investment, acquisitions, and results, drawn from primary sources.

One caveat up front: numbers show the market's heat, and heat does not distinguish the real thing from a relabel. So the last section covers what the numbers do not say, that is, how to tell contract development with "FDE" written on it from the genuine model. If you first want to check whether your own team operates in the FDE pattern, take the AI literacy check.

Jobs: 5,230 percent in sixteen months

Jobs first. According to Indeed data reported by Business Insider in May 2026, postings for Forward Deployed Engineers were 543 percent above January 2025 levels in April 2025 and 5,230 percent above them in April 2026. Year over year, that is roughly a 729 percent increase.¹ The base was small, so the percentages run large, and that should be discounted. Still, almost no other role grows more than fiftyfold in sixteen months.

The movement had begun in the first half of 2025. In June 2025 a16z's Joe Schmidt called the FDE the hottest job in startups and noted that 22 of OpenAI's 311 open roles at the time were FDE or solutions positions.² It was the first sign that the companies building the models had started hiring the people who put them into the field.

Japan moved in 2026. In August, a chaos map organizing the domestic companies that offer FDE-style support was published, sorting 38 companies into seven categories, with each company's start date, presence of case studies, and pricing and contract terms annotated solely on the basis of their own websites and press releases.³ Thirty-eight companies means a market that did not exist a year earlier has entered the stage of comparing vendors.

Investment: $4 billion, $1 billion, and joint ventures with investment firms

Investment next. From spring to summer 2026, the companies that build models and the companies that provide infrastructure put money into deployment. In order:

Date	Who	What	Scale
April 16, 2026	Accenture (Japan)	Dedicated Forward Deployed Engineering practice with Microsoft	Thousands of AI-skilled engineers
May 4, 2026	Anthropic	AI services company with Blackstone, Hellman & Friedman, Goldman Sachs for mid-sized companies	Anthropic applied AI engineers participate
May 4, 2026	Anthropic, FIS	Anthropic applied AI team and FDEs embedded inside FIS to co-design financial-crime agents	Explicit goal of FIS running on its own
May 11, 2026	OpenAI	OpenAI Deployment Company launched, Tomoro acquired	More than $4 billion initial investment, roughly 150 FDEs
June 30, 2026	AWS	New organization embedding FDEs with customers	$1 billion

Accenture's announcement is telling in its wording: the barrier to AI adoption is not a shortage of technology but engineering expertise not being applied on the ground.⁴ A consulting firm declaring a shift from reports to engineering on site.

Anthropic's two announcements came on the same day. One is the joint venture with investment firms, aimed at companies such as community banks, mid-sized manufacturers, and regional health systems that stand to gain from AI but lack the in-house resources to build and run frontier deployments.⁵ The other is the FIS partnership, which states that the point of embedding FDEs is not to stay indefinitely but to transfer knowledge so that FIS can run on its own.⁶

OpenAI's announcement is the largest. More than $4 billion in initial investment, 19 investment and consulting firms led by TPG, and a deployment base of the more than 2,000 businesses those firms sponsor. The investors include consultancies such as Bain, Capgemini, and McKinsey.⁷ The company that builds the models set up a deployment company together with the consulting firms.

AWS's $1 billion matters for one phrase in the announcement: customers leave able to run things themselves, with working systems, documentation, and trained in-house staff.⁸

Acquisitions: buying people

The acquisition numbers are fewer than the investment numbers, but they weigh more. OpenAI acquiring Tomoro to place roughly 150 FDEs from day one shows that the FDE has become a role that cannot be filled by hiring and training alone.⁷ In the same announcement OpenAI says it will keep acquiring firms that can accelerate deployment.

The supply of FDEs cannot keep up because the skills the role requires are not only coding. The skills covered in the first six articles of this series, asking, watching, being there, structuring, filming, handing the dictionary to a machine, are not in an engineering curriculum. Palantir handed new FDEs a user-research manual because it had no choice but to grow those skills in-house.⁹ What takes time to grow gets bought.

Results: the FDE model shows up in margins

Finally, results. Look at Palantir, which has run the FDE model for twenty years. Its second-quarter 2026 results, published in August, show revenue up 93 percent year over year to $1.935 billion, U.S. commercial revenue up 149 percent to $764 million, and GAAP operating income of $912 million, a 47 percent margin. Full-year revenue guidance was raised to between $8.15 billion and $8.16 billion.¹⁰

Read those numbers in the context of the FDE model. A former FDE recalls that around 2016 the company was seen as a Silicon Valley services firm, and that at the time this was not far off, but that by 2023 it had reached an 80 percent gross margin, a software margin rather than the 32 percent of a large consultancy.⁹ The person who led its commercial business cites the public figure of more than $1.1 billion in annual revenue from its top 20 customers and attributes it to deep alignment with those customers.¹¹

Here is one answer to whether an FDE model is real. If learning returns to the product, engineering hours per customer fall and margins approach software levels. If it does not, hours track revenue and margins stay at consulting levels. The numbers reveal the truth of the model, a few years late.

Why 2025 to 2026

Line up the four sets of numbers and the timing coincides. Why then? I see three reasons.

First, the problem of not reaching production was shown in numbers. A report in July 2025 found that despite an estimated $30 to $40 billion in enterprise generative-AI investment, 95 percent of organizations were getting no measurable return, and that the cause was not model performance but the absence of a mechanism to adapt to the context of the field. I covered that report in the article on FDEs versus AI accompaniment. For a company selling models, that is a serious number. If the model sold is not used, there is no next contract. So the decision was made to send people as far as the point of use.

Second, agents began to act. An AI that summarizes text runs even when the company's words are vague. An agent that "allocates orders awaiting approval and issues shipping instructions" cannot run without the company's nouns and verbs. As the ontology article explains, that dictionary can only be built on site. The age of agents is an age that needs someone sitting on the shop floor.

Third, coding agents lowered the cost of code on day one. Until recently only a skilled engineer could ship something that ran on day one, and FDEs were expensive people. Now, given the nouns and verbs, a provisional screen appears in a few hours. The center of gravity of the FDE's work moved from writing to asking, watching, and structuring, and one FDE can cover more sites. I read the slight easing of the supply constraint as one thing that drew the investment.

a16z's Schmidt wrote in June 2025 about trading short-term margin for depth.² With the three reasons in place, 2025 to 2026 was the year that trade became rational.

What the numbers do not say: relabeled FDEs and real ones

This is what I most want to write in this article. When postings rise 5,230 percent and 38 companies line up on a chaos map, most of it is a relabel. Contract-development SE postings rewritten as "FDE"; resident consultants renamed "forward deployed" in a proposal. The numbers do not separate these from the real thing.

The person who led Palantir's commercial business anticipated this in 2024, writing that the imitators copy the form but not the function: they redraw role boundaries, internalize the SI costs they used to outsource, and change how they collect product feedback, without committing to alignment with the customer's interests.¹¹ a16z's Andrusko gave three pressure tests in January 2026: does engineering effort decline on mature accounts, can you say no to customization, and does it still work past 50 customers.¹²

Here are three ways we recommend buyers check. First, ask "where in your product is what you learned at your previous customers?" A concrete answer means the real thing; "we optimize for each client" means contract work. Second, ask "when does it end, and what is left with us when it does?" The answer should be working artifacts, a dictionary, and the ability to fix things yourselves. Third, deliberately raise a branch request and ask "will you decline this?" Whether they can explain their grounds for declining reveals the model.

And those checks point at us too. Since we describe ourselves as working in the FDE style, we owe answers to all three. Learning returns to standard features in ZEROCK and TRAFEED; the end comes in 45 to 90 days; the dictionary's owner sits on site; branches are declined with reasons. This series, including this article, exists to publish those answers in advance.

Summary

The FDE market can be told in numbers: postings up 5,230 percent in sixteen months, more than $4 billion from OpenAI and $1 billion from AWS, a joint venture between Anthropic and investment firms, and Palantir's 93 percent revenue growth with a 47 percent operating margin. The timing converged on 2025 to 2026 because the failure to reach production was shown in numbers, agents began to act, and coding agents lowered the cost of day-one code.

What the numbers do not say is the difference between the real thing and a relabel. To tell them apart, ask where the learning sits in the product, what remains at the end, and whether branch requests get declined. We answer those three too.

If you want to bring the FDE model into your own engineering team, or want an eye for choosing a vendor, take a look at the WARP program or talk to us. The next article covers how FDE outcomes are measured and contracted, and how engineering hours become investment rather than revenue.

Footnotes

1. Job postings for this tech role have grown more than 700% in the last year (Business Insider, updated May 18, 2026, via AOL). The figures of 543 percent above January 2025 in April 2025, 5,230 percent in April 2026, and roughly 729 percent year over year are attributed to Indeed data shared with Business Insider ↩

2. Trading Margin for Moat: Why the Forward Deployed Engineer Is the Hottest Job in Startups (Joe Schmidt, a16z, June 2025) ↩ ↩²

3. Chaos map of FDE vendors: 38 companies in 7 categories for choosing a provider (Marketing Planner Inc., PR TIMES, August 26, 2026, in Japanese) ↩

4. Accenture launches a Forward Deployed Engineering practice with Microsoft to help organizations scale AI across the enterprise (Accenture Japan, April 16, 2026, in Japanese) ↩

5. Anthropic, Blackstone, Hellman & Friedman, and Goldman Sachs launch an enterprise AI services company for mid-sized businesses (Anthropic, May 4, 2026). The description of target companies is my summary ↩

6. FIS Brings Agentic AI to Banking with Anthropic, Starting with Financial Crimes (FIS, May 4, 2026) ↩

7. OpenAI launches the OpenAI Deployment Company (OpenAI, May 11, 2026). The more than $4 billion initial investment, 19 partners, roughly 150 FDEs, more than 2,000 sponsored businesses, and the intention to acquire further firms are from this announcement ↩ ↩²

8. AWS invests $1 billion to embed AI forward deployed engineers with customers (Amazon, June 30, 2026) ↩

9. Reflections on Palantir (Nabeel S. Qureshi, October 15, 2024). The "services company" perception around 2016, the 80 percent gross margin in 2023 against a large consultancy's 32 percent, and the books handed to new hires are from this essay ↩ ↩²

10. Palantir Reports Q2 2026 Results (Palantir Technologies, press release filed as an exhibit to Form 8-K, August 2026). Revenue of $1.935 billion (up 93 percent), U.S. commercial revenue of $764 million (up 149 percent), GAAP operating income of $912 million (47 percent), and full-year guidance are from this document ↩

11. Sorry, that isn't an FDE (Ted Mabrey, September 21, 2024). The more than $1.1 billion from the top 20 customers and the observations about imitators are from this essay ↩ ↩²

12. The Palantirization of everything (Marc Andrusko, a16z, January 16, 2026) ↩