The Forward Deployed Engineer is the thing everyone is talking about. I was at the AI Engineer World's Fair and it was in every hallway. Palantir was there, Replit was there, Anthropic was there, and all of them were talking about FDEs. Postings are up more than 700% in a year. AWS is on it, Replit is on it, Cursor is on it, OpenAI and Google are on it. Every Series A with a real contract wants one.

I have a stake in this. I have built FDE content and training for the better part of a decade, so let me add some detail, because the definition is quietly being hollowed out.

Start with what the term meant when Palantir coined it, somewhere around 2010. Their customers were intelligence and defense agencies, government work, and those customers could not tell you what they needed. The requirements were classified, or they were tacit, or the customer simply had to show you in person. You could not gather them in a conference room and say "send me all your emails and documents." You were confined to the space in question. So Palantir put engineers inside that space to learn by building, in situ, next to the person with the problem.

They called those customer-facing engineers Deltas, a holdover from when Palantir named its business-development teams after the NATO alphabet. Each Delta paired with an Echo, a deployment strategist who owned the human side: institutional politics, workflow, and getting the thing adopted. Turning one customer's bespoke solution into something reusable across many was a separate job, done by Palantir's core product team, the part that paves a gravel road into a product feature. By 2016, Palantir had more Deltas than it had backend developers.

That part of the story gets retold a lot. The part that gets dropped is the platform.

Underneath every bespoke customer deployment sat Palantir's platform, Gotham and later Foundry, doing the undifferentiated work: data integration, access control, the ontology, the plumbing. The Delta's job was the last mile, the bespoke fit to one customer's reality. The platform carried the other nine miles. Take the platform away and that same engineer is doing a full custom build every single time, mile zero to mile ten, which is the exact consulting model the FDE was invented to beat. You cannot drop a customer-facing engineer on-site to start from scratch on every engagement and expect any margin.

We have seen this movie before. SRE started as a real discipline, error budgets, toil reduction, incident command, production ownership, and got flattened into "advanced sysadmin who knows Terraform." DevOps started as a way of working across the wall between dev and ops, and got flattened into "the person who owns the CI/CD pipeline," with no continuous improvement and no feedback flow. FDE is getting flattened right now, and the piece being dropped is the platform.

So look at how the title is actually used in 2026. From the dozen-plus Fortune 100 companies I have talked to about this, there are at least four versions walking around:

  • The Palantir model. Embedded engineer, plus a platform underneath. Bespoke work on a commodity base.
  • The customer-facing consultant. Embedded, writes production code, no platform. Every engagement starts at zero. They might be using Claude, they might get a base handed to them, but they are rebuilding each time.
  • The AI-native engineer on a hyperscaler. Internally facing. Cloud-native and gen-AI skills, deploys on AWS or GCP, but has no commoditized platform of their own and no bespoke agentic software factory. "We want them to do AI-native things, we just are not going to build them a platform to do it on."
  • The in-situ AI engineer with no platform at all. No hyperscaler, no base. Shows up, builds everything bespoke in the customer's infrastructure, and moves on. Holy smokes, that one is hard, and the margin is terrible.

There is a fifth I have seen exactly once out of that dozen: an AI-native engineer with an actual agentic software factory built on a hyperscaler, paved roads and all. One in twelve.

Split each of these into internal and external and it gets messier. Someone asked me this week, in earnest, to turn a room of juniors into amazing consultants with corporate-communication skills who are also fully AI-native, in three weeks. That is the external version, and it is aggressive. The internal version is a little kinder, because you can teach engineering plus some enterprise and solutions design, with real awareness of cost, performance, reliability, security, and operational excellence. The non-functional pillars, which predate anyone's Well-Architected framework by a long way.

None of that changes the core question, which is the platform. The platform is what separates an FDE from a very good, very low-margin consultant. A consultant delivers a bespoke solution that delights the customer, hands it off, and the customer sustains it. They built the whole thing from zero to ten. An FDE delivers a bespoke solution on top of a base that makes the next one faster, and the one after that faster still. Drop the platform and you have not modernized the role. In some cases you have regressed. You have renamed a consultant and lost the compounding.

So when someone tells me they are an FDE in 2026, I am not that interested in whether the label fits. I am interested in one thing: where is your platform? And because the word "FDE" now means six different things depending on who is saying it, do not assume your definition matches theirs. Ask:

  1. 1. Internal facing or external facing?
  2. 2. A software engineer, or a software engineer with AI skills?
  3. 3. Do they have a platform?
  4. 4. On a hyperscaler, or rolling their own?
  5. 5. Does that platform carry and produce AI workloads?
  6. 6. Is there a low-code or no-code agentic software factory in the mix?

If the honest answer to "where is your platform" is that you build everything from scratch in the customer's environment, you are doing real work, but you are doing it the hard way, and you are leaving a lot of margin and efficiency on the table.

Which raises the obvious follow-up. If the platform is what makes the role work and you do not have one, what would it take to build one? That used to be a multi-year, multi-team investment only a Palantir could afford. With the hyperscalers where they are, and platform engineering and IDPs now a known quantity, it is not that anymore. That is the next piece.

And I would be remiss if I did not say it: if you want to learn how to build a platform engineering platform that can host AI workloads, I am teaming up with Packt on July 23 for a four-hour workshop on using Claude to build exactly that, running on AWS. Come on down. The codes are below. Hope you like what you read, and tell me where I have it wrong.

Register: https://www.eventbrite.co.uk/e/agentic-devops-with-claude-build-a-33-component-ai-native-platform-live-tickets-1991969049048?aff=MichaelLinkedIn. Use code MICHAEL40 for 40% off.

Sources

  • Palantir, "Dev versus Delta: Demystifying engineering roles at Palantir" (blog.palantir.com)
  • The Pragmatic Engineer, "Forward Deployed Engineers" (newsletter.pragmaticengineer.com)
  • FDE Academy, "How Palantir Invented the Forward Deployed Engineer Model" (fde.academy)
  • Wikipedia, "Forward Deployed Engineer"
  • FDE posting growth: Indeed data, April 2025 to April 2026