Forward Deployed Engineer

The knowledge is in the building. It just isn't in the software.

A manufacturing company with forty years of history runs on rules nobody ever wrote down. Which supplier you can push on lead times and which one you absolutely cannot. Why that line is never scheduled at full capacity on a Monday. What \"urgent\" means when it comes from that particular client. Which three SKUs you never let go below a certain stock level, and the reason has to do with something that happened in 2011.

None of this is in the ERP. So people route around it — Excel, WhatsApp, and the one person who has been there thirty years and holds the whole model in his head.

This is why manufacturing software keeps failing. The features aren't wrong. The operating rules of the business were simply never encoded anywhere a machine could read them. And the people holding those rules are retiring.

There's a broader shift underway here, most visible so far in professional services: the capital of a firm used to be the expertise sitting inside its senior people. Increasingly, the capital is whether that expertise has been turned into a system. Knowledge that stays in heads is a liability with a retirement date. Knowledge that has been encoded compounds. Manufacturing is the next industry to go through this, and it's the one where the stakes are highest, because the knowledge is more specific, more valuable, and closer to walking out the door.

What we're building

A general-purpose manufacturing OS: the system a company actually runs on, from orders to inventory to production to the active and passive cycle. Under it, a context layer that builds and maintains a living model of how that specific company works — which is what lets an agent act like that company instead of like a generic assistant. On top, agents that do real operational work.

Getting there means solving something nobody has solved: making a 40-year-old manufacturing business legible to software without flattening what makes it work. Every ERP implementation in history has attempted the opposite — force the company into the software's shape and call the difference \"best practice.\" We do it the other way around, and the reason it's now possible is that the cost of building something specific collapsed.

We're not digitizing factories. We're giving them the infrastructure to evolve with AI — turning industrial SMEs into AI-native businesses, so that every capability that ships over the next decade is something they can absorb rather than something happening to other industries.

We've raised €1.7M from Dig Ventures, Vento, 2100 Ventures and others. Customers run their entire operations on us. Italy alone has roughly 650,000 manufacturing companies across 141 industrial districts, and almost none of them have software that fits.

What a Forward Deployed Engineer is

The role was invented at Palantir and has since been adopted by OpenAI and Anthropic, among others. The idea is simple and slightly heretical: instead of building software behind a roadmap and hoping it fits reality, you send the person into reality.

An FDE goes on site. Learns how the business actually runs, from the shop floor up. Then builds the ontology: the formal model of that company's objects, processes and rules — what a \"job\" is here, what states an order moves through, which exceptions are real policy and which are one person's habit. That model is the thing the software runs against. Get it wrong and everything above it is wrong. And you stay until the system is in daily use, which is a very different bar from shipped.

The part that makes it a product role and not a consulting role: everything you learn gets pulled back into the platform. A deployment isn't a project that ends, it's how the product is discovered. FDEs sit upstream of the roadmap, not downstream of it. If what you build stays custom and stays put, the model has failed.

What the role is at Arke

We have two offices: the customer's plant, and ours.

You'll spend real time on factory floors — in Italy and increasingly abroad. You'll sit with production managers, warehouse staff, owners' sons and daughters, and 60-year-old plant directors who are sceptical of you for the first two days and your strongest ally by the second week.

The work, roughly:

• Map how a company actually operates and turn it into the ontology the system runs on
• Translate what operators tell you into specifications, configuration, and working software
• Build and ship — with AI as your multiplier, which is what makes one person able to do what used to take a team
• Handle data migration, integrations, and all the parts nobody puts in the demo
• Train users and drive adoption, which is measured in whether people still use it in month six
• Feed everything you learn back into the platform<