# Munich's Marble Cognition Raises $3.75M for AI Coworkers That Study How Employees Work Before Taking Over
> Marble Cognition raises $3.75M pre-seed led by Heartcore and Seedcamp to build AI coworkers that learn business processes by observing employees.
- **Author:** Prashant Jagtap
- **Published:** 2026-11-02
- **Updated:** 2026-11-02
- **Category:** Startups
- **Editorial type:** news
- **Topics:** agentic-ai, pre-seed-funding, heartcore, seedcamp, munich-startups, enterprise-ai
- **Canonical URL:** https://www.todaysstartupnews.com/startups/marble-cognition-raises-3-75m-pre-seed-ai-coworkers-learn-by-watching
## Key takeaways
- Munich-based Marble Cognition has raised $3.75 million, about 3.35 million euros, in a pre-seed round co-led by Heartcore and Seedcamp
- A desktop application observes how employees work in business software and converts the activity into process maps that AI agents called Marble Coworkers can execute
- Named customers include EGYM, Mutares and Enam, and the company says one US financial firm cut case-handling time by 90 percent
- The money will fund inference scaling, a larger engineering team in Munich and a launch in the US market
Marble Cognition, a Munich-based AI startup that builds software agents capable of learning business processes by observing employees, has raised $3.75 million, roughly 3.35 million euros, in a pre-seed funding round. Heartcore and Seedcamp co-led the round, and the funding was announced alongside the company's public launch on November 2.

Angel investors in the round include a board member of Datadog, a founder of Personio, and executives from McKinsey, Mistral AI and Lovable. Marble has not disclosed a valuation. Co-founder Felix Rösner said the round closed in under two weeks.

The company was founded in late 2025 by Rösner and Christophe Schmit, who serves as chief technology officer. The two met at Munich's Center for Digital Technology and Management. Rösner previously worked at the AI sales-agent company 11x.

Marble is built around a gap that many companies face when they try to automate work with AI. Business processes often remain undocumented, with essential knowledge spread across emails, spreadsheets and the habits of individual employees. That makes it hard to decide which tasks can be automated and how an AI system should carry them out.

Marble points to a November 2025 McKinsey Global Institute report stating that existing technology could in theory automate about 57 percent of work hours in the United States. The company reports that among the businesses it works with, around 90 percent of core processes are not formally documented. According to Rösner, Marble's Coworkers watch how work happens before they take it over.

Rösner has said that many vendors arrive with a predefined use case and a promise of saved hours, without knowing whether the use case is a real problem or how the work is actually done. Marble starts from the current state of a company's operations and builds only where a genuine opportunity exists. In this way the company reverses the traditional automation model, which adapts a standard product to a business process, by concentrating first on understanding the process itself.

The platform relies on a desktop application that observes workflows across business applications, including screen activity, clicks and keystrokes. That information is converted into structured process maps that AI systems can interpret. Marble then builds AI agents called Marble Coworkers, which run the mapped processes from start to finish inside the tools a company already uses. Humans can step in where a task needs judgement or approval.

In one example described by the company, a Marble Coworker at a private equity-backed business prepares comparative offers that a specialist then approves. Marble also says its technology helped a US financial firm reduce case-handling time by 90 percent.

Marble has outlined limits on what the application records. Data is anonymised before any employee or AI model can access it, including inside the customer's own cloud, and it is analysed in aggregate rather than per individual. The software does not monitor private applications, private browser tabs or activity outside working hours, and employees can pause recording at any time. Employees at customer companies can view the analysed data, delete their own data and block specific apps. Marble stores data in Europe, and in Germany for German customers.

Marble says it works with technology unicorns, private equity-backed companies and AI roll-ups. Named customers include EGYM, Mutares and Enam. Rösner said the company also works with large technology firms in Munich and across Germany, private equity firms in Europe and private equity-owned professional services firms in the United States. The stated ambition is to split its customer base evenly between the US and Europe.

Max Niederhofer, a partner at Heartcore, said the scarce input for automation is not the models, which are becoming commodities, but a precise record of how work gets done. Reshma Sohoni, general partner at Seedcamp, said her firm has a history of backing founders who change how people work, citing UiPath and Synthesia, and described Marble as filling the gap of mapping how work is actually performed.

Employee monitoring is regulated in Marble's home market. German law gives works councils a say over technical systems that monitor employee behaviour or performance, and the EU AI Act classifies AI systems designed to monitor and assess employee behaviour and performance as high risk. Rösner described change management as the most difficult part of deployments.

Rösner identifies three groups of competitors: consultancies that map processes manually, vertical AI companies that send engineers into customers to build solutions, and software vendors focused on process mapping. Others pursuing similar budgets include Skan AI, which raised $63 million in a Series C co-led by Cathay Innovation and Dell Technologies Capital, and Munich-based Interloom, which raised $16.5 million in a seed round led by DN Capital in March 2026.

The new capital will be used to scale Marble's inference capabilities, grow its engineering team in Munich and support its launch in the US market.