Here's the dirty secret of enterprise AI adoption: most companies know they need it, almost none of them know how to actually deploy it, and the gap between "we bought a cloud subscription" and "our agents are running in production" is where billion-dollar opportunities are born.

AWS is betting $1 billion that it can fill that gap with a new internal organization dedicated to forward-deployed AI engineers — specialists who physically embed inside client companies to stand up agentic systems and, crucially, leave before they become permanent IT staff. The program was announced Tuesday by AWS VP of Frontier AI Francesca Vasquez, who was careful to frame this as more than a glorified consulting arm.

What "Forward-Deployed" Actually Means

The forward-deployed engineer (FDE) model was popularized by Palantir, which built an entire business around the concept. The idea is straightforward: instead of handing a client a software license and a stack of documentation, you send in your own engineers to embed with the customer's team while the system gets built and stabilized. They learn the client's quirks, they fix things as they break, and — in the ideal version of this story — they eventually leave behind a system the client can actually run themselves.

That last part is what AWS is leaning into hard. The announcement explicitly states that customers should walk away with "lasting AI skills, workflows, and patterns they can use to innovate independently." Translation: the goal isn't to create dependency, it's to create competence. Whether that holds up in practice is a different question, but the framing is notably different from the kind of lock-in strategy you might expect from a cloud provider that profits from every API call.

The model works because AI deployments are genuinely hard in ways that generic cloud tooling can't solve. Every company has different data schemas, different legacy systems, different regulatory constraints, different internal politics about who owns which workflow. A one-size-fits-all agent template doesn't cut it. FDE teams can reuse the core infrastructure across engagements while still tailoring the edges — which is efficient for AWS and actually useful for the client.

The Tradeoff Nobody's Talking About

There's a real cost buried in this model that the press release conveniently skips: maintaining a corps of forward-deployed engineers is expensive and operationally complex. These aren't support ticket handlers — they're senior engineers capable of architecting multi-agent systems, debugging inference pipelines, and navigating enterprise security requirements. Recruiting, training, and deploying enough of them to serve a meaningful client base at scale is a logistical challenge that $1 billion in internal resources only partially addresses.

Worth noting: the $1 billion figure here represents committed internal Amazon resources, not external investment or a joint venture. That's a different structure than what OpenAI and Anthropic have set up.

Everyone's Doing This Now

AWS isn't alone. Both OpenAI and Anthropic have launched their own FDE-style ventures in recent months, valued at $4 billion and $1.5 billion respectively. The key structural difference: those labs partnered with private equity firms, which brought capital and — more importantly — existing portfolios of enterprise clients to sell into. It's a smart shortcut. You don't have to knock on doors if your PE partner already owns the building.

AWS doesn't need that shortcut. It already has the enterprise relationships; the challenge is converting cloud customers into AI-capable operators. The FDE org is essentially AWS admitting that selling compute and managed services isn't enough anymore — you also have to sell competence.

What This Signals for Enterprise AI

The broader pattern here is worth stepping back to appreciate. Three of the most significant AI infrastructure players — Amazon, OpenAI, and Anthropic — have all independently concluded that enterprise AI deployment requires hands-on human expertise that can't be fully productized. That's either a temporary market inefficiency that will disappear as tooling matures, or evidence that deploying AI in complex organizations is fundamentally a sociotechnical problem, not just a technical one.

My read: it's mostly the latter. The technology is capable enough. The bottleneck is organizational — change management, workflow redesign, trust-building between human teams and automated systems. FDE programs are essentially a consulting business dressed in an engineering costume, and there's nothing wrong with that. It's just useful to call it what it is.

The real test for AWS's new org isn't whether it can deploy agents — it's whether those agents are still running six months after the FDE team packs up and leaves. That outcome won't show up in any press release.