Why Anthropic and OpenAI both copied Palantir's model in the same week
In brief
In the same week in May 2026, Anthropic launched a $1.5B deployment venture and OpenAI launched a $10B one. Both are built around a model Palantir pioneered 20 years ago: put engineers inside the client. Here's why the two best AI labs in the world decided to copy a defense contractor.
Contents
In the same week in May 2026, the two most capable AI companies in the world announced they were copying a defense contractor.
Anthropic launched a $1.5B joint venture with Blackstone, Goldman Sachs, and Hellman & Friedman. OpenAI launched "The Deployment Company" — a $10B operation with McKinsey, Bain, and Capgemini — and simultaneously acquired Tomoro, a 150-person forward deployment firm, to staff it immediately.
Both moved within days of each other. The template they are both copying is Palantir's.
What Palantir figured out (and kept quiet for 20 years)
Palantir was founded in 2003 to help intelligence agencies use data they already had. The problem was never the data. The problem was that no one inside the agency had the engineering skills to actually do anything with it — and no outside consultant would stay long enough to build something that lasted.
Palantir's answer: send engineers into the client. Not to consult. Not to deliver a project and leave. To embed — write production code against the client's actual systems, train the people who would maintain it, and build something durable before moving on.
They called these people Forward Deployed Engineers. The model — FDE — was the product as much as the software was.
The result: Palantir built lasting infrastructure inside every major defense agency, several intelligence services, and eventually large commercial enterprises. Their NPS among embedded clients was consistently among the highest in enterprise software. Not because the product was the best, but because someone from Palantir was in the building.
The problem is that Palantir's FDE model was expensive, slow to scale, and deeply tied to their specific software stack. You couldn't separate the model from the product. Which meant no one copied it — until AI changed the math.
Why AI changes the math
AI makes the FDE model dramatically cheaper to scale. Here's why:
The labor cost drops. A Palantir FDE had to deeply understand Gotham/Foundry and be able to customize it at the infrastructure level. An AI FDE's primary lever is knowing how to get Claude (or GPT, or Gemini) to work against a client's specific data, APIs, and workflows. The technical ceiling is lower.
The time-to-value compresses. A Palantir FDE historically needed 6-18 months to build something that changed how a client operated. An AI FDE with access to Managed Agents, MCP servers, and a client's cloud connector stack can ship a working workflow in weeks.
The client side is more ready. Palantir spent much of the 2000s convincing enterprises that software-driven operations were worth pursuing. That argument is already won. Most large companies have decided they need to move on AI. What they lack is an engineer who can translate that intent into a working system.
The compensation is now justified by the value. Senior FDEs at Anthropic's joint venture and OpenAI's Deployment Company are reportedly being offered $300K–$500K+ packages. Companies can afford to pay this because the workflows these engineers unlock are worth multiples of that in productivity gain.
What the two ventures actually are
Anthropic's joint venture
Anthropic partnered with three of the most significant capital allocators in the world — Blackstone (the world's largest private equity firm), Goldman Sachs, and Hellman & Friedman. These partners don't just bring money: they bring their portfolio companies as the first set of clients.
The structure is a separate entity from Anthropic, with dedicated engineering teams. The engineers go into Blackstone's portfolio companies, Goldman's banking operations, H&F's investments. They build systems against real enterprise environments using Claude, Cowork, Managed Agents, and the full Anthropic stack.
Anthropic's upstream benefit: every deployment is a production feedback signal. Every system an FDE builds informs what Claude needs to do better in enterprise contexts. This is the same reason Palantir invested so heavily in FDEs — the field teams were also their best product researchers.
OpenAI's Deployment Company
OpenAI moved even more aggressively. They launched a majority-owned entity with McKinsey, Bain, and Capgemini — the three consulting firms most synonymous with enterprise transformation — and immediately acquired Tomoro to bring 150 FDEs in on day one.
The McKinsey/Bain/Capgemini partnership is notable. These firms have the enterprise relationships and the project management infrastructure. OpenAI has the model. The FDEs are the delivery layer that connects the two. It's vertical integration of the go-to-market stack.
What this means if you're building AI systems for companies
Two conclusions are clear from this:
1. The Palantir model has been validated at scale. When the two most valuable AI companies in the world both invest billions in the same go-to-market approach, that's not a trend. That's the playbook. If you're building or managing AI systems for companies — whether you're an aspiring FDE, a working Agent Operator, or an IT director evaluating vendors — this is the structure you'll be operating inside.
2. Consulting firms are now deployment infrastructure. McKinsey, Bain, Accenture, Deloitte, KPMG, EY — every major firm has either launched an FDE practice or partnered directly with an AI lab. The gap between "strategy consulting" and "AI implementation" is closing fast. What used to be a 12-month McKinsey engagement to recommend a software architecture is becoming a 6-week FDE sprint to build it.
The honest filter
The simultaneous launch of two multi-billion-dollar deployment ventures in the same week is the clearest possible signal that forward deployment is the commercial model for enterprise AI.
But this is worth saying plainly: neither Anthropic nor OpenAI is building something that will compete with the engineers inside their own companies. The FDE ventures are the delivery channel. The model quality is still what determines whether the systems FDEs build actually work.
If you're considering a career path toward forward deployment — see How to become a forward deployed engineer for the actual path.
If you're an IT director evaluating whether to bring in an FDE vs. building internal capacity — the answer depends entirely on your org's technical depth and how specific your AI use cases are. The Agent Operator vs. FDE distinction is the right frame for that decision.
Try this today
If you work at a company that's been slow on AI adoption, forward this article to your CIO or VP of Engineering with one line: "This is the model the market is moving toward. We should have a conversation about how we want to participate — as buyers, builders, or both."
That conversation is happening inside every large company right now. Being the person who starts it is the first move in either the FDE path or the Agent Operator path.