Marble raises US$3.75 million for AI agents that learn workplace tasks
The funding will support a larger Munich engineering team, expanded inference capabilities and a US market launch.
Marble Cognition has secured US$3.75 million in pre-seed financing to develop AI agents that learn how employees work before taking on business tasks. Heartcore and Seedcamp jointly led the round, with angel investors also participating.
Felix Rösner and Christophe Schmit founded the company in late 2025. Its technology is intended to turn employees’ day-to-day activities into structured process maps, giving AI systems instructions for carrying out work that may not otherwise be documented.
Marble says around 90 per cent of core processes at companies it works with lack formal documentation. Its agents, called Marble Coworkers, are designed to use an organisation’s existing tools, with people remaining involved where judgement or approval is needed.
A desktop application records screen activity, clicks and keystrokes across workplace software. Marble says it anonymises the information before staff or AI models can access it and analyses activity collectively, rather than employee by employee. It says private apps, private browser tabs and activity beyond working hours are excluded, and employees can pause observation whenever they choose.
The company says organisations including EGYM, Mutares and Enam already use its technology. It plans to spend the funding on expanding inference capabilities, adding engineers in Munich and supporting its entry into the US market.
