OpenAI is building a whole company around it. Anthropic is doing the same. Google, Mistral, and Cohere are recruiting as fast as they can. The job listings have spiked over the past year. The title is the same everywhere: Forward Deployed Engineer. It is a sought-after role in AI right now. But the concept itself is not new.
Palantir invented the role almost twenty years ago
The term comes from Palantir. Early on, they faced an unusual problem. Their customers in US intelligence could not explain what they needed. The requirements were classified. So were the processes. Palantir’s fix was straightforward: they moved the engineers out to the customer. The engineers sat in the customer’s environment. They saw the problems first hand. They built the solution right where it would be used, taking responsibility all the way from the first needs analysis to operations. This model became so central that Palantir long had more Forward Deployed Engineers than regular software engineers.
The rest of the industry treated it as a curiosity. It was expensive. It was hard to scale. And it was tightly coupled to Palantir’s own platform. The consulting world simply carried on as usual, with pre-studies, reports, and handovers in PowerPoint.
Then came AI, and the model fit
Generative AI has a known production problem. The models ace the demo. But faced with real data, undocumented workflows, and existing systems, most initiatives stall out. An MIT study from 2025 reviewed hundreds of AI pilots at large enterprises. The result? Roughly 95 percent produced no measurable impact on the bottom line. The researchers’ conclusion was telling: the fault was not in the models. It was in the integration.
This is exactly the problem a Forward Deployed Engineer solves. An engineer who sits in the customer’s team sees how the work actually gets done. They find the processes that are actually worth automating. They build, evaluate, and deploy in the very environment where the solution will live. A requirements specification does not beat sitting next to the person doing the job today.
The AI companies have drawn the same conclusion. In May 2026, OpenAI launched The Deployment Company, armed with over four billion dollars in capital and a single mission: to take the models all the way into enterprise operations. Days earlier, Anthropic presented a similar initiative with Blackstone and Hellman & Friedman. The model companies have realised a simple truth: models do not deploy themselves.
Platform or ownership
There is a nuance here that is easy to miss. At Palantir, the role existed to make customers succeed with Palantir’s own platform. The end goal was always to land the business in the vendor’s product.
Now, the way of working has been decoupled from the platform. When AI needs to be adapted and integrated into a real business, the embedded model works best, regardless of whose models are used. The next question then becomes decisive: who owns what has been built when the engineers go home?
How we work
For us at TokenTek, this is just how we work. It has been that way from the start. Our engineers work at the customer, in the customer’s teams, and within their infrastructure. We train the customer’s people as the solution takes shape. The difference from the platform model is what happens at the end of the journey. When we hand over, the customer owns everything. The models, the code, the documentation, and the way of working.
The industry has put a name on how we work. We take that as a compliment.
Want to see what it means in practice? Get in touch and we will show you.