The forward deployed engineer (FDE) phenomenon has moved front and center in conversations with everyone from start-up founders to listed tech firms to Wall Street analysts. FDE has become a proxy for a promise of certainty. And for those who have spent the best part of three years failing to scale AI in the enterprise, that’s quite the lure, complete with the promise of business outcomes delivered.
That FDE promise also includes going deep on how work gets done in your organisation (aka ‘context’ – or ontology in the Palantir canon), and committing to the hard yards of making the AI work in your context, then sticking with you to keep it working, increasingly more efficiently, once it’s in production.
That’s the whole appeal. As HFS CEO Phil Fersht put it: “FDE isn’t a new job category. It’s the market’s response to three years of AI pilots that nobody was accountable for turning into outcomes.” Match that promise and enterprise leaders will care very little about the job title, the mechanics, or the team behind it.
With 88% of enterprises still stuck short of scaled AI value (HFS AI maturity research), that promise is intoxicating. Which is why it needs qualifying.
The qualification test runs in both directions
HFS spoke to Ed Kerr, Head of Product, Software and Services at Rackspace Technology, to understand the Rackspace-Palantir model; in which Palantir supplies the context fabric and Rackspace supplies the sovereign runtime, the systems engineering, and the L1 FDE bench.
Kerr is blunt about who this is not for. Ask a chief AI officer what they’re doing with AI and get back “we use a lot of ChatGPT to help us write better documents”? And his response would be: “Shut down the Palantir conversation. I’m not trying to build you a better chatbot.”
A Palantir-shaped problem, in his words, isn’t a localised workflow. It’s stringing 50 agents across 14 departments into an end-to-end enterprise workflow — and it’s usually existential. “If I don’t solve this problem quickly, you’re going to lose your job.”
The economics enforce the discipline. A platform of the kind required to do the heavy lifting runs to tens of millions of dollars of hardware and meaningfully hosts around ten customers. Scarcity flips the sales conversation: prove to me your problem is worth one of my ten spots. Rackspace builds the POC before charging a penny, then takes a fraction of the value released. Tellingly, a customer who has already tried and failed to solve the problem is a positive signal – it proves the problem is serious and the commitment is real. This is not for beginners.
The same scale test shows up in the tokenomics. Dedicated GPUs beat frontier API pricing only at high, sustained enterprise utilisation. Below that line, you’re better off on the API — and you were never a Palantir-shaped problem in the first place.
The Bottom Line: FDE is an accountability model, not absolution
FDE won’t save your chatbot project, and you shouldn’t expect it to. It closes the distance between AI capability and enterprise context — but it is only worth the journey if you have context worth capturing and the outcome is worth the money.
Before you go looking for someone to absorb your risk, do Kerr’s filtering for him. Is your problem enterprise-spanning, existential, and expensive enough to justify the certainty you’re buying?
