When Palantir invented Forward Deployed Engineering (FDE) in the 2010s, the industry mocked them as "consultants with equity." Nobody's laughing now.
During the opening keynote, @swyx called out the brand new FDE track at the AI Engineer World’s Fair as one of the things he was most excited about. Cursor just hired a VP of Forward Deployed Engineering. Anthropic teaches 101 classes on it. LinkedIn says it's the fastest-growing job AI has created, with postings up 42x since 2023. In May, OpenAI put $4B behind DeployCo, an entire company built out of FDEs. At AIE, nine companies explained how they embed engineers with customers, and half ended with hiring pitches. When that many companies are hiring for the same role, it's time to pay attention.
AI products fail at integration, not awareness. Everyone at this conference knows Cursor exists. The question is whether it works in your janky monorepo. That's a problem DevRel can't solve. Awareness is already won, and another blog post won't get you past the integration wall. Someone has to get into the codebase and make it work. That's the FDE.
What is a Forward Deployed Engineer (FDE)?
An FDE is the vendor's engineer, embedded in the customer's codebase, shipping production code. They sit with the customer's team, learn the weird edge cases the demo glossed over, and build the actual integration. Their code merges into the customer's repo and stays there.
The customer pays, through services line items or contracts priced to include deployment. DevRel is a cost center under marketing, but FDEs sit on the revenue side.
It's easy to confuse with roles you already know:
- Not a sales engineer: FDEs stay post-sale and build real systems.
- Not a consultant: they only deploy their own product.
- Not support: they have commit access.
Nat Meurer, Head of Agent Engineering at Sierra, made the argument at AIE that AI software is moving to outcome-based pricing, and someone has to guarantee the outcome. That someone has commit access, not a content calendar.
So who makes a great FDE? Engineers with customer empathy, ex-consultants who ship, sales engineers who'd rather build, senior product engineers bored of internal roadmaps. Communication is the differentiator, not raw coding. AI narrowed that gap.
If I were hiring FDEs today, I'd look in two places. First, hackathons: people who build under pressure with strangers are the archetype. Second, your own DevRel team: they've been doing the empathy half of this job all along, and if you're doing DevRel right, they're already highly technical.
So is DevRel dead?
No. But the funnel is splitting. A billion new software creators are coming online, and their agents will increasingly be the ones making buying decisions. Winning those software creators and the agents choosing tools on their behalf is still DevRel's responsibility.
DevRel owns the top of the funnel: awareness, winning the customer's attention. FDE owns the bottom: converting that attention into real usage.
What gets squeezed is the middle. Developer marketing aimed at engineers who read docs and deliberate is the most at risk. That audience is shrinking from both ends. The disposable-software crowd delegates the decision to agents and the enterprise crowd wants an FDE in the room.
There's a credibility inversion here: A conference talk earns applause, but a merged PR in the customer's repo earns a renewal.
DevRel isn't dead. Somebody still has to win the attention of a billion new software creators and the agents buying tools on their behalf. But attention alone never made an AI product work in production. That's the piece we've been missing, and it's the piece FDE fills. The funnel finally has both halves. Go ship in someone else's repo.
Top comments (2)
A blog post is written once and then works while you sleep, for thousands of readers at a time. An FDE works one codebase at a time and then goes home. That gap is the thing I'd add to your funnel: the bottom half you describe is real, but it costs headcount that scales in a straight line, while the top scales for free. I actually wonder if that's why giants can field armies of FDEs and a solo tool can't, which quietly turns your split into an edge for exactly the big players. Does the outcome-based pricing really cover an engineer sitting in each customer's repo, or is it being subsidized right now to win the logos?
the complementary framing is right, but i'd name the mechanism under it: FDE is the market repricing the gap between "sounds credible" and "survives my codebase". devrel sells the claim, the FDE ships the evidence, and for a long time that gap was small enough that a good blog post covered it. with AI products it isn't. the demo works and prod drifts, so the distance between the claim and the evidence got wide enough that someone has to physically walk it into the customer's repo.
that's also why the merged-PR line lands so hard. a blog post asserts the thing works, a PR in the customer's codebase is the one artifact that can't be faked, because it had to survive contact with their actual mess. one is a claim, one is evidence, and every renewal you lose is hiding in the space between the two.
so it isn't devrel dying, it's the integration truth you used to be able to defer. the buyer discovered the friction in month three and marketing had time to paper over it. now an engineer or an agent tests in-repo in hour one, the last mile shows up immediately, and someone has to actually be standing in the codebase when it does.