The AI vendors want to move into your building

The pitch changed. AI vendors stopped selling you a tool and walking away. Now they want to send their own engineers into your building to build the thing, run the thing, and stick around. In the last few weeks, four of the biggest names in AI all made the same move within days of each other. Let me catch you up on what that means for your team.

Everyone landed on the same play

The model is called forward-deployed engineering. A vendor sends its own engineers to sit inside your company, figure out where AI actually helps, and build it into your workflows. Palantir pioneered it two decades ago. Now it's the hot thing in enterprise AI.

The money arrived fast. Amazon committed $1 billion to an internal FDE org on a Tuesday. Two days later, Microsoft announced a $2.5 billion "Frontier Company" with 6,000 people. Some inside Microsoft suspect Amazon caught wind of the plan and rushed to go first. OpenAI and Anthropic already launched their own versions, backed by TPG, Goldman Sachs, and Blackstone.

Microsoft's Judson Althoff wouldn't even use the FDE label. He called it "the largest, most capable, outcome-driven engineering organization in the industry". Strip the branding and it's the same bet everyone else is making.

Why they're all doing it now

The demos worked. The rollouts didn't. Businesses bought ChatGPT, Claude, and Copilot, then found that impressive demos don't turn into results inside a real company with its own data and rules. "Having the model alone doesn't change your workflows or how you operate," Goldman Sachs' Marc Nachmann said about the Anthropic deal. "You need people who can combine the technology with what's actually happening in the business."

There's a business reason underneath it too. AI models are becoming commodities, cheaper and more alike by the month. The real money is in the services that make AI pay off, a far bigger market than selling the models. So the vendors are chasing it.

A startup named Trase put a number on the problem. Its president, an ex-AWS engineering leader, says AI adoption "is faltering within sectors that need it most". His line: the issue isn't innovation, it's implementation.

The lock-in hiding in the handshake

Microsoft is selling swappable models. Satya Nadella wrote that no company should "cede value to a few models that eat everything they see", and pitched the ability to trade one model for another as the test of whether you still control your future. Sounds good.

Watch the gap between the pitch and the wiring. Even if you can swap the model, letting Microsoft's engineers build your systems means those systems end up running on Microsoft's cloud and Microsoft's tools. Jumping ship gets very hard. You trade model lock-in for platform lock-in, and the second kind is stickier.

AWS is at least honest that its version wants you self-sufficient. It promises customers leave "with both new solutions and new engineering capabilities", plus skills and patterns they can reuse alone. Ask any vendor to put that in writing.

How new is this, really

Microsoft already runs a big in-house delivery arm, thousands of consultants and engineers who build inside customer companies. It has FastTrack. It has a dedicated practice with Accenture and a $1 billion alliance with EY. The Frontier Company is a bigger, better-branded push behind work the company was already doing. Consulting roles were even expected to be hit in a coming layoff round the same week.

That matters for how you read the sales call. You're not being offered something brand new. You're being offered old services delivery with an AI label and a bigger price tag. Judge it as delivery work: who owns the code, who keeps the knowledge, what happens when they leave.

The deep cut

The fight underneath all this is over who owns your institutional knowledge. Nadella's whole argument is that you should keep it. But the FDE model works by embedding a vendor's people so deep in your workflows that the knowledge lives in their heads and their platform, not yours. That's the real risk, and it doesn't show up in the demo.

So before you sign, make the exit plan the first conversation, not the last. Write down what your team keeps when the engineers pack up: the code in your repo, the documented workflows, the trained people. If a vendor won't commit to leaving you self-sufficient, you're not buying capability. You're renting dependence. Note that Nadella's model-agnostic pitch is also the pitch from smaller players like Neo and Trase, so you have leverage to demand it.

Three questions for your team

1. If we let a vendor's engineers build our AI systems, what exactly do we own and keep on the day they walk out? Get it in the contract.
2. Are we buying model flexibility that's real, or model flexibility that's fake because everything still runs on one company's cloud?
3. Where is our AI actually stalling, in the tech or in the rollout? If it's the rollout, an embedded team helps. If it's the tech, they can't fix that.