On 11 May 2026, the five of us founded LumeSec Technologies. Here is what we keep watching break in process automation, what we started this company to fix, and where it begins.
A note from day one
On 11 May 2026, Christoph Piminger, Manuel Granbichler, Christian Sternig, Felix Diem and I founded LumeSec Technologies GmbH. This is the first thing I want to put under that name: not an announcement, just a short note about why we started this and where it begins.
We did not come at this from outside. We have spent our careers inside regulated finance and enterprise IT — sometimes on the building side, sometimes on the risk side, sometimes as the people the regulator sat across from. Between us we have seen the same gap appear in too many regulated environments.
What we keep watching break
A team buys a workflow tool, or builds an agent, or wires an RPA script — and within months the work is back on a human's desk. Something upstream changed, or a step touched a system it shouldn't have, or nobody can explain what the thing did when an auditor asks.
The model is rarely the problem. The system around the model is — no shared memory of what happened, no boundary on what each step is allowed to do, no way to replay a run when something goes wrong. That is the gap we want to close.
What we are building
An operating layer underneath process automation, so the work actually completes — predictably, repeatedly, without a person hovering over every step. We call it AIOP, the AI Operating Platform. It rests on three primitives:
- Correlate — every signal and every action joined into a single replayable timeline.
- Contain — every agent bound to an identity and scoped to what it may touch.
- Attest — every consequential step signed by a named human or service.
We don't want "AI inside your tools." We want processes that finish themselves, with the operator in the loop where it matters and not where it doesn't.
The AI Operating Platform
The operating layer underneath the AI you run — agents, identity, policy, evidence under one contract.
See the platformDocument Understanding
Hundreds of pages in, structured data out — replay-equivalent extractions on every field.
How it readsFinancial Document Analysis
Balance sheets in, defensible analysis out — source reference on every figure.
Numbers you can defendWhere we start
From June 2026, our first design partners come onto the platform. We start with one real process per partner — contained scope, measurable outcome — because that is how production-grade automation gets built: prove it on real work, then scale from there.
If you have a process that should run end to end without a person nudging it — not a chatbot, not a side-panel copilot — write to me at marcus@lumesec.ai. I read everything that lands there.
That is the company on day one. The next sprint is already underway.
— Marcus, with Christoph, Manuel, Christian and Felix