Forward Deployment Is Not a Pure Tech Role: From Palantir to a Ten-Step Path for Enterprise AI
Many organisations say they want to adopt AI. The real difficulty is rarely another Copilot licence. It is finding people willing to walk into the company, face incomplete procedures, reluctant departments, conflicting emotions, and relentless KPIs, and put AI into everyday work.
That role is the Forward Deployed Engineer (FDE).
We believe it is still relatively new, extremely hard, and genuinely in demand. And the person who leads it does not always have to be the strongest coder in the room.
Where FDE came from
FDE is not a job title invented in 2026.
Palantir was among the first to turn it into a working method. They sent engineers into government, intelligence, and defence environments instead of shipping software remotely and asking clients to install it themselves. The reason was straightforward: processes were hard to write down in advance, data could not simply leave the site, and requirements changed every day. People had to be present: observe, then build.
In the generative AI era, model companies hit the same wall: the models are strong, but enterprises cannot get value out of them. OpenAI therefore created The OpenAI Deployment Company (also known externally as DeployCo): teams that go on site to help organisations build AI applications and embed them in real work.
One line captures the arc:
- Palantir proved that people must go in.
- OpenAI proved that even model companies must own deployment.
What is missing in the middle is not another chatbot. It is people who know how to enter the field.
Why the role feels new, and why it is so hard
FDE is hard because it asks for several capabilities that rarely sit in the same person.
First, it is dynamic.
No two companies share the same processes. A method that works at Company A may be useless at Company B the next day. You need AI literacy, domain knowledge, an agile mindset, and the ability to adapt in real time.
Second, you face people, not only systems.
Inside organisations there is often defensiveness or outright resistance. Departments may not cooperate. Some fear replacement; some fear failure; some simply protect their own KPIs. An FDE spends the day negotiating, explaining, and compromising, while still delivering measurable outcomes.
Third, results are challenged.
“We used AI” is not enough. Stakeholders ask: How many hours did we save? Did headcount really come down? Did resources fall? Stories are easy; measurement is hard.
Yet demand is real. Almost every enterprise wants AI, but most stall at trials, proofs of concept, or internal contests, and never reach day-to-day operations.
So we hold a view that is not mainstream, but serious:
This role does not always have to be led by IT or a pure technical background.
Sometimes what you need more is judgement about people, communication, and the ability to get things done inside an organisation.
Technical skills can be learned or staffed. If you cannot handle people and departments after you arrive, projects die at “the system is done; nobody uses it.”
Our method: ten steps to bring AI into the enterprise
We do not treat forward deployment as a one-off consulting report. We treat it as a repeatable loop.
Understand how the whole company works, and what SOPs already exist. Do not push tools before you have seen the real workflow.
Talk to different stakeholders. Engage leadership, middle management, the front line, IT, and compliance. Each sees a different pain.
Map systems and AI the company already has. Many organisations are not starting from zero. They may already have Copilot, a CRM, RPA, or an internal generative tool. The problem is often shallow, fragmented use.
See how far existing software actually goes in practice. Feature lists and real usage usually sit far apart.
Use design thinking to find work that is both easy and worth doing. Do not open by attacking the most sacred, political, or difficult process.
Use a four-quadrant lens to decide what to do first, and what not to do. Two axes: easy ↔ hard; needs human judgement and warmth ↔ needs repetitive process.
- Easy + repetitive: first priority. AI prototypes show hour savings fastest.
- Hard + repetitive: do later, when capacity exists; break down and govern.
- Easy + human / warmth: AI as assistant. Do not replace judgement or relationships.
- Hard + human / warmth: do not automate first; people must stay in the room.
Start in the first quadrant (easy + repetitive). That is where you prove AI is not a stage show.
Use AI prototypes to reinforce existing work, not to rip out the stack. Improve and embed first. Do not begin by rebuilding the platform.
Measure improvement. At minimum: hours saved, headcount impact, resource use. Without numbers, the next budget round rarely arrives.
Collect user feedback, then iterate. If the front line finds it hard to use, even a polished demo is only an internal performance.
Carry the learning forward and open the next pain point. The value of forward deployment is not finishing one process and leaving. It is helping the organisation learn to continue on its own.
Three judgements for enterprises
No two companies share the same processes.
So every engagement is customised. The only question is how much: some need existing tools wired well; others need a workflow redesigned. Selling a “standard package” is easy. On site, it almost always bends.
Companies should build their own ACE Team.
That is an AI Capability & Enablement Team.
External forward deployment can lead the way, complete the first loop, and leave a method. Long-term capability must stay inside the company.
Why internal staff often make the best backbone for that team
They already understand the business, so you skip a layer of “explaining the company to outsiders.” They often face less supplier-style compliance friction. Most importantly: colleagues helping colleagues usually meet less resistance than “outsiders rewriting my job.”
The external FDE’s role is to deliver the first success and help the ACE Team keep the method, not to stay forever as a substitute for them.
Closing
From Palantir’s work with government departments to OpenAI building its own Deployment Company, the message is the same:
The bottleneck in AI has moved from “is the model strong enough?” to “is anyone on site making it work?”
The role feels new because it is dynamic, difficult, and entangled with people and KPIs. Demand exists because almost every enterprise wants AI. We believe the people who succeed need more than AI knowledge: they need judgement about people, communication, and the ability to move one concrete thing forward inside imperfect organisations.
Understand how the place works. Talk first. Use the four quadrants to pick easy, repetitive work. Prove savings with a prototype. Then hand the loop to the company’s own ACE Team.
This is not a one-time implementation. It is how an enterprise turns AI into organisational capability.
If you want to see how we practise this with Hong Kong teams, read our Forward Deployed Engineering (FDE) approach.
Ready to talk?
If you or your organisation has questions about AI adoption, AI training, or AI implementation, we are ready to help. We would be pleased to hold a short briefing.
