Marble Cognition raises $3.75M pre-seed to build AI coworkers from observed workflows

Marble’s approach focuses on capturing how employees perform tasks across applications, translating those observations into structured process maps, and deploying AI coworkers to execute workflows within existing tools. The startup claims this method bridges the gap between understanding workflows and automating them, addressing a key challenge: most companies lack formal documentation of their operational procedures. Early results from a financial-services deployment show a 90% reduction in handling time per case and a 50% increase in case volume without additional hires. However, these figures are company-reported and specific to one deployment, not a general benchmark.

Marble emphasizes privacy controls, such as excluding private applications and anonymizing captured data, to address concerns about observing employee activity. The company’s desktop application records screens, clicks, and keystrokes within business tools, but employees can pause recording, and analysis is performed anonymously. The startup’s goal is to enable AI agents to handle repetitive, tool-heavy administrative tasks while preserving human judgment for exceptions.