Every AI platform today is built on LLMs that give different answers each time. That's fine for writing marketing copy or to summarize a report. But it fails in high-risk industrial environments where a wrong output could trigger a regulatory violation, halt a production line, or fail a financial audit.
This gap in AI orchestration exists because many of the people building them come from the same world - SaaS companies, or cloud native enterprises. In these environments, an imperfect or inaccurate AI output is an inconvenience, but doesn’t have real consequences. Of the thousands of vendors currently claiming to offer agentic AI, Gartner estimates only around 130 are genuine.
But genuine AI orchestration for production environments requires deterministic outputs, on-premise deployment, and governance that regulators accept. INXM is built with all three in mind.
The Insight That Built INXM
Robotic Process Automation (RPA), software that handles repetitive, rule-based tasks, promised to automate enterprise workflows. But because RPA depends on systems staying exactly the same, a single change could bring the whole process down. A bot reading data from SAP runs without problems until SAP updates its interface, moves the field two pixels to the left, and the bot breaks. Companies ended up with teams of people whose entire job was fixing broken bots. Systems that can't handle change don't last in environments that change constantly.
Alexander Oelling saw this play out at Isar Aerospace, where he was Chief Digital Officer. He spent years digitising production workflows to launch orbital rockets, with zero room for failure. A launch abort triggered by unreliable software costs tens of millions of dollars and potentially a revoked operating license. In that environment, brittle automation is a liability. He’d watched the same problem go unsolved for years: knowledge workers still copy-pasting between SAP, Excel, and email to get their daily work done.
When agentic AI arrived, Alexander looked to solve the problems around what happens when systems runs in a real environment, every day, without a safety net.
He knew what a system for that environment had to look like. He brought together Matthias Kainer, who'd built the launch orchestration systems at Isar, Jesper Bylund, who'd designed flight control interfaces and Kamil Klüber, who'd spent seven years at Siemens selling into manufacturing.
Production-Grade AI Orchestration
INXM compiles AI into fixed, reproducible workflows. It runs on the enterprise's own infrastructure, not external cloud servers. Cloud automation tools can't reach software on a factory's private servers; INXM can, because it runs there too. The data never leaves the building, keeping it compliant with GDPR and the AI Act.
Organisations design each process once, then lock it into a reproducible workflow with precise logs, audit trails, and human sign-off on any critical action. “Reliable enterprise software is usually painful to use. I’ve made it our priority that INXM isn't”, says CPO Jesper Bylund, former Head of Design at n8n.
Proving It Works
INXM's first production deployment showed what deterministic orchestration could do. At one early customer, factory managers used to spend 40 to 70 manual steps producing a single report - opening SAP, extracting data, reformatting in spreadsheets, cross-referencing MES outputs, chasing departments for sign-off. With INXM, this now takes under 10 steps, and you can trust the system to run identically every single day without variance.
Manufacturing is the wedge because it's where unreliable AI costs the most. A production report that varies from one day to the next doesn't just annoy a manager, it halts lines, fails audits, and corrupts the systems everything else depends on. That's where deterministic execution stops being a nice-to-have and becomes the only acceptable option.
But the same problem exists across every industry that runs on systems that can't afford to be wrong.
INXM is making progress top-down, as it doesn't need an engineering team to run. Someone in the business describes a process in plain language. INXM turns that description into a Plan: a fixed, auditable workflow that does the same thing every time. A manager approves it once, and it runs reliably from then on.
This is what "compiled AI" means in practice. The language model does the thinking once, up front, to build the workflow. After that, the Plan runs like software, not like a chatbot guessing each time. It sits on top of the company's existing systems - ERP, CRM, and the rest - and coordinates between them without replacing anything. Because governance, audit trails, and human approval are built in, a company can go from describing a process to running it in under thirty days.
Why We're Backing Them
At Cherry, we see a huge opportunity to back founders building for Europe's industrial economy.
We estimate the addressable European market for industrial orchestration at $10 to $15 billion, a slice of the $116 to $155 billion that industrial enterprises spend on IT each year. Manufacturing accounts for 20% of European GDP. But the category is much bigger than one geography or sector. For example, every bank or energy company faces the same constraints: they need AI orchestration that works reliably in production, not just in demos.
As INXM moves into energy, and financial services, the addressable market multiplies. As the US and UK tighten data sovereignty requirements, the category becomes global.
Enterprise AI is stuck in a paradox: the more ambitious the deployment, the less predictable the outcome. INXM built the way out. We’re proud to back the infrastructure layer regulated industries have been waiting for.
