The best origin story in enterprise software is about internal tooling. Palantir’s forward-deployed engineers kept solving the same integration problems customer after customer, until the product engineers watching them began encoding the recurrences — ontology, permissioning, provenance — as reusable primitives. That layer became Foundry. Nabeel Qureshi, who lived inside the loop, describes it plainly: the field did bespoke work, the platform team mined it for patterns, and the patterns became the product.
Gergely Orosz’s history makes the scale of that field era concrete: until 2016, Palantir employed more forward-deployed engineers than conventional ones. The platform was not designed in a lab and carried to the field. The field came first, and the platform condensed out of it.
Today, everyone running a forward-deployed motion is rebuilding that layer. Sierra decided agents need their own software development life cycle and built one — regression suites assembled from annotated real conversations, release gating, an audit surface. Distyl, founded by ex-Palantir engineers and reported by The Information at a $1.8 billion valuation, ships its engineers with the “Distillery,” a platform for turning organizational knowledge into AI workflows. OpenAI split its field force outright: forward-deployed engineers deliver customer systems, while a separate group — forward-deployed software engineers — builds the abstractions those systems are made from. An internal platform team, for the field.
Different companies, different verticals, one convergence. Everyone in forward deployment ends up building the same piece of software. And everyone builds it themselves, from scratch, in private.
The convergent architecture
Look at what the layer always contains, whoever builds it. A way to capture the client’s operating reality — systems, constraints, approval chains — as something structured, rather than as tribal knowledge in one engineer’s head. A way to gate privileged actions, because embedded engineers touch production and someone has to be answerable for that. A way to retain evidence of what was done and why, because someone will eventually ask. A way to transfer ownership, because the engagement ends. Capture, gate, retain, transfer. The architecture converges because the job is identical everywhere: place engineering authority inside someone else’s operation without losing control of it.
Palantir has now taken the logic to its endpoint. The “AI FDE” shipping inside Foundry is an agent that does forward-deployed work — closed-loop execution, branch-and-propose changes, human review before merge. The company that invented the job title is converting the job into software. That is what conviction about this layer looks like after more than a decade of accumulating it.
Why it never becomes a product
And yet outside the labs and the billion-dollar verticals, the layer gets built the same way every time: internally, once, under deadline, by whoever had a spare sprint between engagements. It never gets productized, because the people who need it most are fully consumed by the delivery work it is supposed to support. There is a rational-sounding reason at every step — the client is waiting, the pattern is not quite general yet, the engineer who understands it is mid-engagement — and the sum of the rational reasons is that the firm’s most valuable intellectual property lives in Slack threads and muscle memory.
The serious objection is that this is exactly how it should be — the internal platform is the moat, so why would Sierra or Distyl ever license theirs? For them, the objection holds. But the forward-deployed motion is no longer five companies. It is every AI consultancy, every systems integrator standing up an applied-AI practice, every product company hiring its first field engineers because the enterprise deals demand it. OpenAI’s willingness to fund a second engineering organization purely to build field abstractions tells you what this layer costs to do properly; almost nobody below the labs can staff that. For everyone else the real choice is not build versus buy. It is buy versus operate forever without one — every engagement starting from zero, every pattern living in a departing engineer’s head.
The category has no vendor, which is strange, because the evidence that it wants to exist is a decade deep and gets restated every time another field team builds it again. The method is licensable. The memory is the asset.
That is the category LockedIn Labs FDE occupies: the platform layer under forward deployment, built as a product a practice can run its own method on — memory included.
LockedIn Labs FDE is the platform forward-deployed engineers carry into the enterprise — the operating reality held as a governed asset, workflows and agents as reviewable definitions, and every privileged action stopped at a named human.