The Forward Deployed Engineer, Chapter 19: The Future of Forward Deployment

This is the final post in the series walking through my book The Forward Deployed Engineer. In the previous chapter, we covered hiring. Chapter 19 closes the book with the question most readers ask first: what happens to the role next.

I have, throughout the book, treated the Forward Deployed Engineer as a stable operating pattern. Chapter 19 admits the obvious complication. Nothing about AI in 2026 is stable, and the role is changing under its own feet. What I’ve tried to do in the final chapter is not predict where the function lands — predictions are cheap and usually wrong — but to frame the four shifts already visible inside the function as of this writing, so that the reader who picks the book up in 2027 can recognize what changed and decide which of my conclusions still hold.

The most consequential shift is the gradual automation of parts of the role itself. Agentic tooling now does the kind of integration work an early-career FDE used to do, and it does the work faster than a human can review the output. Field engagements in 2026 use coding agents to draft customer-specific integrations that, three years ago, would have been a junior FDE’s first month of billable work. Eval harnesses are increasingly generated by agents that watch the workflow and propose the metrics that should govern it. Even some of the discovery work — the workflow inventory in particular — is starting to be assisted by agents that observe the customer’s operations directly and produce a first-pass inventory the FDE then corrects. The role is being absorbed unevenly, from the bottom up, and the early-career rungs are disappearing fastest. This has career-design implications I’ll get to in a moment.

What I want to be precise about is what the automation wave cannot, in my view, replace. The political work of stakeholder mapping — building a map by name, role, motivation, and risk — is not automatable, because it requires reading rooms and judging people on signals agents do not have access to. The trust-building work with the Skeptic is not automatable, because trust is a relationship between humans and the Skeptic does not extend it to systems. The strategic work of deciding which workflow to redesign, given a customer’s political and operational reality, is not automatable, because it requires the kind of judgment that emerges from years of seeing similar deployments succeed and fail. And the judgment of when to push back on a customer’s stated requirement — when to tell a Chief Risk Officer that the deployment they’ve asked for is the wrong one — is not automatable, because it requires standing behind the pushback with one’s professional reputation. These are operator capabilities, not technical capabilities, and they remain the load-bearing part of the role.

FDE-as-a-Service, the Talent Market, and the Customer Wave

Two longer-term trends deserve their own treatment. The first is FDE-as-a-Service: the emergence of specialized firms that supply trained FDEs into companies that don’t want to build the function themselves. The model is, in my view, going to be one of the most important organizational stories in enterprise AI over the next three years. Some companies will hire FDEaaS firms because they cannot build the function quickly enough. Some will hire them because they have concluded, correctly, that they shouldn’t build it at all. And some will hire them as a hybrid alongside their own function, using the external team for surge capacity and industry-specific specialization. Each of these is a different operational story, and the chapter walks them in detail.

The second trend is the long-term evolution of the FDE talent market itself. For most of the role’s history, FDEs were engineers who rotated through the role on the way to something else. In 2026, for the first time, I see a meaningful number of senior engineers explicitly building their careers around the role — choosing it as their long-term home rather than as a tour of duty. This is healthy. It is also what makes the early-career disappearance I mentioned earlier survivable. The career path is now clear enough that engineers can join the function as career FDEs rather than rotators, and the function can compensate them accordingly.

The fourth shift, and perhaps the most important, is that the customer base for FDE engagements is itself growing faster than the supply of FDEs. Industries that were not buying enterprise AI two years ago are now buying it in volume. Industries that bought it cautiously are buying it aggressively. The function’s longer-term maturity will depend less on what the labs that invented the role do than on what the customers who learn to host it do — and the chapter walks which industries are leading, which are lagging, and what changes when an industry develops its own FDE-equivalent counterpart roles on the customer side.

This is the last post in the blog series. Nineteen chapters, nineteen days, one core idea: the operating model is the asset. Models change. Platforms change. Customer industries change. The discipline of bridging the generalizable platform to a specific customer’s operations — the discipline the FDE practices — does not. If you’ve read along over the past three weeks, thank you. If you want the whole system in one place — every framework, every case study, every appendix, the full DARE treatment, the full Eval-Customer Split, the full OLA dashboard build, and the operating model that holds it all together — the book is where it fits together.