Palantir’s forward-deployed engineering model was, for years, the thing enterprise software companies pointed to as an example of what not to do. Too labor-intensive. Too hard to scale. Too dependent on elite engineers being physically embedded in customer environments. Every SaaS wave of the 2010s was built on the opposite premise: self-serve, product-led, low-touch.
That premise held for workflow software. It does not hold for enterprise AI.
Why FDE is back
The failure mode of enterprise AI is predictable: the model works in the demo, breaks in the customer’s environment, and dies in committee while the implementation team waits for IT security approval. That sequence happens because no AI product ships with the integration logic, data cleaning, change management process, and employee training required to actually operate in a specific customer’s stack.
FDE programs solve this by putting engineers inside the customer. Not salespeople. Not solutions architects. Engineers who can see the actual data, understand the actual workflow, and write the integration code that makes the agent work in this environment, not a generic one.
This is expensive. It’s also, apparently, the only thing that works at scale for complex enterprise AI deployments. ServiceNow + Accenture’s program this week is the clearest mainstream validation of the model. Cognizant launched a parallel version on May 7th. Several others are following.
Which FDE programs are real
Not all FDE programs are created equal. The signal to look for: ratio of engineers to account managers. A real FDE program has engineers outnumbering AMs. A marketing FDE program has one engineer paired with four sales people and a lot of slide decks about “embedded expertise.”
The second signal: pre-built agent skills. FDE programs that have to build every workflow from scratch for every customer are not scalable. The ones that will win have a library of validated, production-tested agent modules — call it the “skill catalog” — that the FDE team deploys and customizes. The customization is where the human time goes; the plumbing is pre-built.
ServiceNow’s announcement specifically called out 300+ pre-built agent skills. That number is worth paying attention to. It suggests a real investment in reusable deployment infrastructure, not just a marketing wrapper around consulting hours.
Implications for smaller vendors
The hard truth for Series A and B enterprise AI companies: you probably can’t afford a true FDE program. The margins don’t work at your scale. What you can do is architect your product so that your largest customers can run their own lightweight version of it — a configuration layer deep enough that a customer’s internal team can adapt the agent without needing your engineers on-site. That’s the product design challenge. Most teams are not thinking about it early enough.
The alternative is to partner with system integrators who run FDE programs — get your product into Accenture’s skill catalog, Cognizant’s Secure AI Services offering, IBM’s Consulting arm. That distribution motion is underutilized by most AI startups. The SIs are looking for product. The startups are looking for scale. The match is obvious.