Marble Cognition raises $3.75M to build AI coworkers that learn by watching employees work

Marble Cognition has raised $3.75 million in a pre-seed funding round co-led by Heartcore and Seedcamp, Seedcamp announced on November 2, 2026. The Munich startup builds AI coworkers that learn a job by watching employees do it, rather than by reading documentation about it.

The problem Marble is selling against is undocumented work. Founder Felix Rösner estimates that 90% of the core processes at the companies he has worked with live solely in inboxes and in people’s heads. Asking employees to describe their own work, Seedcamp argues, produces a pared-back account that misses the edge cases, missing information and delays that shape the real process.

Watch first, automate second#

Marble’s desktop app observes how teams work, recording the screen, clicks and keystrokes in the business apps people use at work. It turns that observation into a structured map of the process that an AI system can read like code, then builds Marble Coworkers that run the process end to end in the tools teams already use, with humans in the loop where judgement matters.

“Every company we talk to wants AI to take over real work. But nobody can tell the AI how that work is actually done,” Rösner said. “So we solve that first. Our Coworkers watch how the work happens before they take it over. They do it the way your best people do.”

The privacy line#

Observation is the sensitive part of the pitch, and Marble’s answer is built into the product’s limits. Everything it captures is anonymised before any person or AI model sees it, even in the customer’s own cloud, and data is only analysed in aggregate, not based on individuals. The app never records private apps, private tabs or anything outside working hours, and employees can pause it at any time.

Early results, and who is already using it#

The earliest evidence is a single customer story, reported by Seedcamp rather than independently verified. At a private-equity-owned US financial services company, a Marble Coworker cut handling time per case by 90%. In the first month, the team handled 50% more cases without new hires. In both measures, Seedcamp says, the AI was taught by observing, rather than through documentation.

Tech.eu reports the technology has already been deployed at several organisations, including EGYM, Mutares and Enam, as well as other technology companies and private-equity-backed businesses. Marble has not published a wider customer list, and the 90% and 50% figures describe one company’s first month, not a controlled comparison.

What the money is for#

The funding will be used to scale inference, grow the engineering team in Munich and launch in the US. The round was co-led by Heartcore and Seedcamp, with angel participation from a Datadog board member, a Personio founder, and executives from McKinsey, Mistral AI and Lovable. Marble was founded in late 2025 by Rösner and Christophe Schmit.

Seedcamp frames the bet plainly: the technology to automate processes already exists, and what is missing is an exact picture of how the work gets done. Marble is productising that understanding itself. Whether observation-first automation earns trust inside companies that have to watch it work will depend on the privacy limits holding up in practice, and on results beyond a single early customer.