The Forward Deployed Engineer Goes Mainstream: Selling AI Now Means Embedding

What Happened

On June 30, AWS said it was standing up a dedicated Forward Deployed Engineer (FDE) organization and putting $1 billion behind it. Francessca Vasquez, AWS’s VP of frontier AI engineering and services, led the announcement. An FDE is an engineer who works from inside the customer’s building, writing and wiring the AI agents that solve that specific company’s problems. AWS plans to seed the org with “thousands” of these engineers, sending pods of roughly five or six into a single customer at a time, where they work alongside AI agents. The stated goal is fast engagements and customer self-sufficiency: “Customers leave AWS FDE deployments with both new solutions and new engineering capabilities,” Vasquez said.

What matters is that this isn’t only an Amazon move. OpenAI earlier spun up a $4 billion FDE joint venture, and Anthropic a $1.5 billion one. Both paired with private equity firms, not just for capital but for a ready list of portfolio companies to sell into. Amazon went the other way. Instead of carving out a separate entity, it built the org with internal resources, which keeps the IP in-house, wires it tightly into the AWS stack, and gives customers a cleaner “one vendor, one P&L” story. None of this is new, though. Palantir invented the model in the early 2010s, when intelligence-agency customers couldn’t articulate what they needed, so Palantir stationed engineers on-site. By around 2016, FDEs outnumbered its traditional software engineers.

What This Means for Founders

Three companies bet billions on the same idea in the same week. The signal is hard to miss: selling AI to the enterprise is shifting from “ship a product” to “embed an engineer.” Why now? Because models that shine in the lab routinely break once they hit a real customer, messy production data, decades-old legacy systems, business rules nobody ever wrote down. A demo can’t survive that. Right now the only reliable way to close the gap is to put a human inside it. The AI labs have simply rediscovered the problem Palantir solved twenty years ago.

For founders, the reframe is the point. The Valley instinct has been self-serve, product-led growth: ship, let customers onboard themselves, keep humans out of the loop to protect margins. FDE inverts that. Embedding your best engineer in the customer isn’t a tax on the P&L; it’s your distribution channel, and often the only one that makes the product actually work in production. Y Combinator has told early founders the same thing for years: do things that don’t scale, sit next to your users. What changed is that the largest AI companies now treat it as core strategy, not a phase you grow out of. So the fork is here. On one side, teams that read field deployment as pure headcount cost. On the other, teams that turn each engagement’s repeatable patterns, templates, connectors, internal tooling, back into product. That is exactly why Amazon kept the IP in-house. Every deployment has to compound into an asset that makes the next one cheaper.

What You Can Do Now

If you sell AI to enterprises, stop treating founder-and-early-engineer time inside a customer as overhead and start treating it as your first distribution channel. But don’t stop at one-off customization. Every integration you repeat, every data-cleaning step you redo, every prompt you re-tune, capture it as a template or an internal tool so the next customer takes half the time. And nail down IP ownership of that shared tooling in the contract from day one. What Palantir and Amazon show is that the moat isn’t putting people on-site; it’s turning that fieldwork back into owned, reusable assets. That one clause is often what decides whether your third customer is finally profitable.