# Marble Cognition raises $3.75M to have AI watch how you work, then do it

> The Munich startup records clicks, keystrokes and screenshots to map real workflows so its AI coworkers can take them over.

- URL: https://dev.devisdead.ai/news/marble-cognition-raises-3-75m-to-have-ai-watch-how-you-work-then-do-it
- Published: 2026-11-02T13:02:07Z
- Author: DevIsDead Newsroom
- Section: Startups & Money
- Tags: agents, automation, funding, workplace
- Originally reported by SiliconANGLE (https://siliconangle.com/2026/11/02/marble-cognition-raises-3-75m-for-its-ai-agent-workers-that-watch-and-learn-how-people-work/).

The pitch is simple and a little unsettling: before an AI can do your job, it has to watch you do it.

Munich-based Marble Cognition launched with a $3.75 million pre-seed round led by Heartcore and Seedcamp, as [SiliconANGLE](https://siliconangle.com/2026/11/02/marble-cognition-raises-3-75m-for-its-ai-agent-workers-that-watch-and-learn-how-people-work/)'s Mike Wheatley reported. Angels include executives from McKinsey and Mistral AI and a Datadog board member.

## What it does

Co-founders Felix Rösner and Christophe Schmit argue that agents are already everywhere but lack the context to automate real work, because most processes live in email threads, spreadsheets and people's heads. Marble's desktop app observes employees, capturing screenshots, clicks and keystrokes, and turns that into a structured process map that agents can follow almost like code. Its agents, which it calls Coworkers, then run the repetitive parts end to end across existing software, with people kept in the loop for important decisions.

On privacy, Marble says recordings are anonymized and can be reviewed before any model sees them, processing can stay in the customer's private cloud or on premises, and it filters out personal browsing.

## The numbers it claims

Early customers include EGYM, Mutares and Enam. At one private-equity-backed US financial services firm, Coworkers drafted settlement offers for human review, which the company says cut case handling time by 90 percent and let staff handle about 50 percent more claims without new hires.

## The reality check

Those are vendor numbers, and process mining is not new; UiPath built a business on similar ideas. What's new is the end goal: not a dashboard of how work flows, but an agent that takes it over. For the people being recorded, the question isn't only privacy. It's what happens to the job once the map is complete, and whether they had a say in drawing it.

**[Scoreboard](/scoreboard) update:** AI +1. Agents learn jobs by watching the people who do them.

Stay undead.