# The forward deployed engineer land grab

Source: https://tpiros.dev/blog/forward-deployed-engineer-land-grab

Between 4 May and 2 July 2026, four companies committed roughly $9 billion to the same idea. OpenAI launched [the Deployment Company](https://openai.com/index/openai-launches-the-deployment-company/) with $4 billion from a TPG-led investor group, acquiring the UK consultancy Tomoro and its ~150 engineers in the process. Anthropic introduced [Ode](https://www.anthropic.com/news/enterprise-ai-services-company), a $1.5 billion joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs. AWS stood up a [$1 billion forward deployed engineering org](https://www.aboutamazon.com/news/aws/aws-1-billion-forward-deployed-ai-engineers). Microsoft announced [Frontier Company](https://www.cnbc.com/2026/07/02/microsoft-commits-2point5-billion-6000-employees-ai-implementation-unit.html): $2.5 billion and around 6,000 people.

The idea they're all funding: embed their own engineers inside your company to make AI work there.

The job title for this is forward deployed engineer, and it's having a moment. [Indeed listings grew from 643 to 5,330](https://en.wikipedia.org/wiki/Forward_Deployed_Engineer) between April 2025 and April 2026. If you run engineering at a normal-sized company, some version of "should we get FDEs?" has probably already reached you from the board.

Before it does (or again, after it has), it's worth being precise about what this role costs and who it's for.

## What a forward deployed engineer actually is

Palantir invented the role. An FDE embeds with the customer and writes production code against the customer's real data and systems, owning a business outcome rather than a slide deck. That last part is the whole difference from consulting: a consultant recommends, an FDE ships. By 2016 Palantir reportedly employed more FDEs than product engineers.

The reason demand exploded in 2025 is well documented. MIT's Project NANDA found that [95% of enterprise GenAI pilots showed no measurable P&L impact](https://www.forbes.com/sites/janakirammsv/2026/05/28/ai-giants-bet-billions-on-the-most-expensive-job-in-enterprise/). Companies bought models and got demos. The gap between a demo and a system that survives contact with production data and actual users turned out to be enormous, and FDEs exist to close it. The demand is rational. [a16z called it the hottest job in startups](https://a16z.com/services-led-growth/) back in June 2025, and for once the label stuck because the underlying problem is real.

## What one costs, and who gets one

Here's where it stops being relevant to most companies.

When OpenAI ran FDEs as an internal service, The Information reported a minimum customer spend of [$10 million](https://www.business-standard.com/companies/news/openai-custom-ai-consulting-service-10-million-grab-accenture-125070200681_1.html). Early engagements included the US Department of Defense at roughly $200 million, and Grab. Senior lab FDEs earn $350,000 to $550,000 a year, so the arithmetic can't work any other way.

Look at how the new ventures are structured and the target market gets even clearer. Ode's investors are Blackstone, Hellman & Friedman, Apollo, General Atlantic. The Deployment Company's are TPG, Advent, Bain Capital, Brookfield. These joint ventures are built to sell into their investors' portfolio companies. That's the distribution model: private equity introduces the vendor to a few thousand large enterprises it already owns.

One mid-market provider put it plainly: the labs [will not start sending humans](https://www.utsubo.com/blog/forward-deployed-engineer-studio) to smaller companies. If your organisation is 100 to 2,000 people, no lab is embedding engineers with you at any price you'd sign.

What you'll be offered instead is the label. Deloitte now has a [forward deployed engineering service page](https://www.deloitte.com/us/en/services/consulting/services/forward-deployed-engineering.html). Staffing shops rent "fractional FDEs" from about $2,500 a month. Some of the downmarket pods are probably good. But Palantir's acceptance rate for the role was reportedly 6 to 10%, and you don't fill 6,000 seats at that bar. [The dilution critique](https://seattledataguy.substack.com/p/forward-deployed-engineering-is-about) writes itself: much of what now wears the FDE badge is implementation consulting with a 2026 haircut.

## The incentive question

Suppose you are big enough for the real thing. There's still a structural issue worth naming before you sign.

A forward deployed engineer is the vendor's employee. Every one of those four billion-dollar vehicles exists to grow consumption of that vendor's models or cloud. The engineer embedded with your team can be excellent and personally honest, and the arrangement still points one direction: deeper integration with the vendor who sent them. Anaplan's CEO called the model [an effective sales tactic but a poor long-term strategy](https://www.forbes.com/sites/stevebanker/2026/07/10/palantir-and-forward-deployed-engineering-what-should-we-believe/) for the buyer, because it leads to lock-in.

The vendors know buyers have noticed. AWS now markets its FDE programme on customers being [self-sufficient when a deployment ends](https://www.aboutamazon.com/news/aws/aws-1-billion-forward-deployed-ai-engineers). When the exit shows up in the sales copy, it's because procurement teams started asking what's left behind.

That's the right question, so ask it formally. If you're negotiating an FDE engagement, write the exit into the contract:

- Your engineers pair on the work from week one, not shadowing in the last fortnight.
- The evaluation suite (the tests that say the system still works) is handed over, documented, and runnable by your team.
- Named internal owners for every deployed component before the engagement ends.
- A defined self-sufficiency check: your team ships a change to the system, unassisted, before the vendor rolls off.

If a vendor resists putting capability transfer in writing, you've learned what the engagement is really for.

## The version of this that fits everyone else

Here's the number I keep coming back to. An [analysis of 1,000 FDE job postings](https://bloomberry.com/blog/i-analyzed-1000-forward-deployed-engineer-jobs-what-i-learned/) found that 58% were at companies of 11 to 200 people. Companies that size are trying to hire a title invented for $10 million enterprise engagements.

Strip the title off and look at what the role does all day: ship LLM features against messy internal data, build retrieval over the company's documents, wire models to real tools with guardrails, write evals so anyone can tell whether the thing works. None of that is arcane. It's a skillset, and working engineers pick it up in weeks when they build against their own systems rather than watching slides about someone else's.

That's the honest comparison for a mid-market company. Renting the capability means it arrives fast, costs a lot, and leaves with the vendor. Growing it in your own engineers is slower to start, an order of magnitude cheaper, and it stays. The $9 billion currently being spent on the first option says a lot about how valuable the underlying capability is. It says nothing about which way of acquiring it fits a 400-person company.

I have an obvious interest here: teaching engineering teams to do exactly this is [what I do](/programmes). Discount accordingly. But the argument stands on the public numbers, and the cheapest time to check it is before the FDE pitch reaches your board, not after.