The most valuable person at an AI company right now might not be the researcher training the next model. It’s the engineer who shows up at a customer’s office and makes that model actually do something useful. The Information reports that forward deployed engineers, or FDEs, have become one of the most sought-after roles across the AI industry, and the reason says a lot about where the money is moving.
What stands out here is the shift in what’s scarce. A year ago, the bottleneck was raw model capability. Now the models are good enough for most business problems. The hard part is bending them to a specific company’s messy data, weird workflows, and legacy systems. That’s the FDE’s job.
What an FDE actually does
The role borrows directly from Palantir, which built its business on engineers who embed with clients instead of shipping software over the wall. The playbook is now spreading fast:
- They sit inside the customer’s operation, not at headquarters.
- They build custom integrations, prompts, and tooling on top of the core product.
- They feed what they learn back into the product roadmap.
- They turn a stalled pilot into a signed contract.
Think of them as a hybrid of sales engineer, consultant, and product builder. Less “here’s the API, good luck” and more “I’ll be at your desk Monday.”
Why it matters now
Foundation models are commoditizing. OpenAI, Anthropic, and Google all sell access to capabilities that look increasingly similar from the outside. When the model is no longer the differentiator, deployment becomes the moat. The company that gets a customer to real production value first is the one that keeps the account.
That’s why FDEs are suddenly everywhere. They’re the bridge between an impressive demo and revenue that actually lands. For AI labs racing to justify their valuations, that bridge is the whole game.
There’s a second reason this is significant. FDEs generate something money can’t easily buy: ground truth on how AI breaks in the real world. Every failed integration, every hallucination in a production workflow, every edge case becomes signal that improves the product. The companies with engineers in the field learn faster than the ones watching from a dashboard.
The trade-off nobody loves
This model has a catch. FDEs don’t scale like software. Each one can only be in so many places. Heavy reliance on them starts to look less like a high-margin software company and more like a consulting firm, which investors reward with much lower multiples.
The bet most AI companies are making: use FDEs to crack the hard early deployments, learn what’s repeatable, then productize it so the next customer needs less hand-holding. Whether they can actually climb down from that high-touch model is the open question. Palantir, worth noting, still leans on it years later.
What to do about it
If you run or work at an AI company:
- Treat deployment as a product, not an afterthought. The gap between demo and production is where deals die.
- Hire or train FDEs before you think you need them. They’re your fastest path to learning what customers actually do with your tech.
- Watch the ratio. If headcount in the field grows faster than revenue per customer, you’re building a consultancy, not a platform.
If you’re a buyer of AI tools, ask vendors who’s going to sit with your team during rollout. The answer separates the companies that will get you to value from the ones that will leave you with an API key and a prayer.
If you’re an engineer weighing a career move, the FDE path is one of the rare seats that touches customers, product, and revenue at once. That combination tends to age well.
The broader signal is hard to miss. The AI industry is entering its delivery phase. Building the model was the last war. Getting it to work inside real businesses is the next one, and the people who can do that are about to be very well paid. More details are available in the original reporting from The Information.