The Forward Deployed Engineer, Chapter 12: The New AI Wave — OpenAI, Anthropic, Runway, Greptile

This is Part 12 of a series walking through my book The Forward Deployed Engineer. In the previous chapter, we tore down Palantir. This one surveys the new wave — four AI labs that have rebuilt the function for the agentic era.

For most of the 2010s, the Palantir operating model was a curiosity. Few outside the company understood it; fewer still attempted to copy it. From 2023 onward, that changed completely. The model became the template. It is, I would argue, the most important organizational story in software since the rise of the platform PM — a defense-intelligence operating model, reborn at the frontier AI labs, with each lab adapting it to a different customer surface and a different commercial reality. The four I’ll survey here are not exhaustive, but they cover the spread.

OpenAI’s enterprise function has grown rapidly since the company’s 2020 commercial pivot. Where it has settled, broadly, is a structure of FDE specializations by industry, deep partnerships with a small number of design partners, and an unusually clean split of eval ownership between the FDE team (customer-specific evals, workflow evals, deployment evals) and the platform research function (model evals, safety evals, capability evals). The split is not always tidy in practice — the boundary moves quarterly — but the existence of the split, and the discipline of negotiating it explicitly, is one of the things OpenAI’s function does that founders building adjacent FDE teams should study.

Anthropic’s enterprise function has grown around its commercial Claude offering and the Claude Agent SDK, and its structural choices are different in interesting ways. The function is tightly integrated with the company’s safety and alignment research, which produces an unusually high research-engineering bar inside the FDE team itself. The team’s operating cadence runs a deliberate customer-research feedback loop — weekly synthesis of field engagements into research questions, monthly handoff of those questions to the research function — that I have not seen run quite as systematically anywhere else. The pattern is worth watching for any founder whose product depends on the model itself improving alongside the deployment.

Runway and Greptile, and the Constants Beneath

Two specialized labs round out the survey, and both stretch the role’s edges in instructive ways. Runway serves film studios, advertising agencies, and creative production houses — a customer base whose workflow-integration challenges look nothing like financial services and whose tolerance for governance ceremony is, for better and worse, quite different. The FDE at Runway has to be fluent in the language of post-production pipelines, not procurement and risk committees, and the operating model the company has built around that customer surface is a sharp adaptation of the Palantir template to a creative-tools customer that approaches AI tools more as collaborators than as services.

Greptile sits at the other extreme. Its customers are large engineering organizations adopting AI-assisted code search and development, which means the FDE has to embed inside another engineering team. The deployment is, in a real sense, into the customer’s own developer experience, and the soft-stack moves required to land that — how you persuade a senior engineering team to adopt a tool that changes how they work — are a study in their own right. The pattern Greptile has developed for this kind of embedded engagement is worth study for any founder whose customer base is itself technical.

Taken together, the four labs make a striking case for how much of the underlying operating model survived the transition from Palantir to the new wave. The customer surface shifted from defense to commercial. The commercial maturity arrived faster (these are SaaS-like sales cycles, not three-year defense procurements). The model-frontier proximity is closer, because the FDE is now deploying systems whose underlying capabilities are still moving quickly. But the deployment-as-product framing, the senior-IC staffing, the operating model as the asset — all of it survived. The chapter closes with three patterns I’ve seen emerging across the new wave that founders should track: the rise of vertical FDE pods organized around customer industry; the formalization of eval-as-product, with evals shipped to customers as part of the deployment surface; and the FDE-as-research-engineer hybrid that is now competing with applied research teams for the same hiring pool. Tomorrow, the failure modes.