Marble Cognition Lands $3.75M to Teach AI Agents How Employees Actually Work

While artificial intelligence agents have proliferated across enterprises, many organizations struggle to deploy them effectively. Marble Cognition, a newly launched startup, aims to bridge this gap after securing $3.75 million in pre-seed funding. Heartcore and Seedcamp led the round, with participation from angel investors including executives at McKinsey & Co., Mistral AI and a Datadog Inc. board member.

Co-founders Felix Rösner and Christophe Schmit, German AI engineers based in Munich, recognized that while AI agents have become commonplace, most companies lack the operational context needed to deploy them effectively. Organizations typically use AI agents for software development, but realizing their full potential requires understanding the granular details of what needs to be built and automated.

The fundamental obstacle lies in how business processes actually operate. McKinsey research indicates that 57% of U.S. work hours could theoretically be automated, yet most companies fall far short of this potential. More than 90% of business processes remain embedded in email chains, scattered spreadsheets and employee knowledge rather than documented procedures.

Marble's solution involves capturing workflows directly from employee activity. The company has built a desktop application that monitors how workers use various tools and processes, recording screenshots, clicks and keystrokes. This data transforms into structured process maps that AI agents can interpret and execute, functioning similarly to code.

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

Felix Rösner

Privacy considerations receive attention in Marble's design. The company anonymizes all recorded worker data and allows review before it reaches any AI models. Processing occurs within customers' private cloud or on-premises infrastructure. The system also identifies non-work-related computer activity, such as social media browsing during breaks, and excludes it from collection.

After gathering data, Marble deploys its Coworkers—AI agents designed to handle repetitive processes end-to-end across existing software systems, with humans making critical decisions. Early customers including EGYM GmbH, Mutares SE & Co. KGaA and Enam GmbH have reported significant efficiency improvements. One private equity-backed U.S. financial services firm configured Marble's Coworkers to generate settlement offers for human review, reducing case-handling times by 90% and enabling staff to process approximately 50% more claims without additional hiring.

Reshma Sohoni, General Partner at Seedcamp and an early investor in robotic process automation firm UiPath Inc., emphasized the market opportunity. "The vast majority of companies in the world want to learn how to automate more of their work," Sohoni said. "The hard part has always been mapping out how the work gets done, so you can actually automate it. Marble fills that gap, giving companies vital context for AI transformations that could define their future."