Marble Cognition raises $3.75M to map workflows for AI agent workers

Marble Cognition, a Munich startup co-founded by Felix Rösner and Christophe Schmit, launched today with a $3.75 million pre-seed round. Heartcore and Seedcamp led the round, and angel investors including executives from McKinsey & Co., Mistral AI and a board member of Datadog also took part. The company argues that most enterprises already use AI agents but lack the operational detail needed to tell them what to automate.

Marble's approach is a desktop application that monitors how employees use their tools, recording screenshots, clicks and keystrokes. That data is turned into a structured process map that agents can read and execute. The company says the data is anonymized and can be reviewed before it reaches any AI model, and that processing can run in a customer's private cloud or on-premises, with non-work browsing excluded.

Its Coworker agents take over repetitive processes end-to-end across existing software, with humans kept in the loop for key decisions. Rösner says early adopters include EGYM GmbH, Mutares SE & Co. KGaA and Enam GmbH. At one U.S. financial services firm, the agents drafted settlement offers for human review, and the company reports case-handling times fell by 90%.

Developers building agents for enterprise work may find that capturing how a process runs today is the bottleneck, rather than writing workflow definitions by hand. Marble's model implies a desktop-monitoring data pipeline, so teams planning something similar should check the anonymization and on-premises options first. The efficiency figures are company-reported results from early customers, not independent measurements.