Marble Cognition raises $3.75M to turn hidden work into AI automation

Marble Cognition, an AI startup that builds digital assistants by watching employees perform their jobs, has raised $3.75 million in a pre-seed round co-led by Heartcore and Seedcamp, with angel backers including a Datadog board member, a founder of HR software firm Personio, and executives from McKinsey. The funding will accelerate Marble’s core technology—a desktop tool that records and analyzes workflows across business applications—while expanding its engineering team in Munich and preparing for a U.S. launch.

The investment reflects growing interest in AI that learns from observation rather than predefined rules. Unlike tools that require manual process mapping, Marble’s system starts with raw data: screen activity, clicks, and keystrokes. Its AI then reconstructs those interactions into executable workflows, creating agents that replicate human performance without needing explicit instructions.

The approach aligns with research showing that more than half of U.S. working hours could be automated, but most companies lack the resources to document their processes. Marble’s co-founders, Felix Rösner and Christophe Schmit, put it simply: *"No one can tell AI how work is actually done. So we solve that first."* The startup’s "Coworkers" aren’t just observers; they’re designed to handle tasks once they’ve learned them, working alongside existing tools while allowing human oversight.

Privacy and adoption could still be hurdles. Marble stresses that data is anonymized, aggregated, and limited to work hours—employees can pause monitoring anytime, and private applications or browser tabs are excluded. Still, the idea of AI passively recording workflows might face resistance in organizations concerned about surveillance-like practices. Similar concerns have arisen with other automation tools, though those typically rely on explicit scripting rather than passive learning.

The funding comes as AI automation startups experiment with different strategies to stand out. Earlier this year, we covered Supermemory AI, which focuses on retaining contextual information from emails and documents—a related but separate challenge. Marble’s focus, by contrast, is on the "shadow work" that most automation tools ignore. If it succeeds, the model could attract mid-sized companies frustrated by the complexity of traditional automation or the lack of precision in generative AI agents.

Key questions moving forward: Can it handle real-world complexity? Marble’s claim that it can analyze workflows without manual input will be tested as it expands beyond early adopters. The $3.75 million will fund engineering to improve its analysis, but whether those agents can adapt across industries remains unclear. Will it gain traction in the U.S.? The presence of McKinsey-affiliated investors suggests confidence in the market, but adoption will depend on showing clear benefits compared to simpler tools. How will competitors respond? Startups in process mining or AI-native automation may see Marble as either a competitor or a potential collaborator. If its "watch-then-automate" approach gains momentum, it could push others to refine their methods—or risk being left behind.

The round highlights a shift in AI automation: it’s no longer just about replacing repetitive tasks. It’s about understanding how those tasks are actually performed—a change that could redefine productivity tools in the years ahead. For businesses, the question isn’t whether AI will automate more work, but how quickly it will stop needing detailed instructions to do so.

Sources: tech.eu