Marble raises $3.75M to build AI coworkers that learn by watching

Solving the documentation gap

Many companies face a significant hurdle when trying to implement artificial intelligence: they do not actually know how their own internal work gets done. While technology exists to automate many tasks, the specific steps are often buried in individual emails, spreadsheets, and the mental knowledge of long-term employees. This lack of clear instructions makes it difficult for organizations to figure out which activities are actually ready for automation.

Marble Cognition was founded to address this specific bottleneck by identifying and understanding these undocumented workflows. The startup notes that roughly 90 per cent of core business processes in the companies they work with are not formally written down. This creates a massive barrier for any business trying to transition from manual labor to automated systems.

The company argues that while many businesses want AI to handle real work, they cannot provide the necessary blueprints for that work to happen. By focusing on the discovery phase first, Marble aims to create a structured foundation that other AI systems can then use to perform tasks accurately. This approach moves the focus from simply having an AI tool to having an AI that actually understands the specific context of a company's operations.

How Marble Coworkers learn

The core technology behind Marble involves a desktop application that observes how employees interact with their computers. It tracks various inputs such as screen activity, clicks, and keystrokes across different business applications. By watching these actions, the platform can begin to build structured process maps that represent the flow of a specific job or task.

Once these maps are created, the system develops what the company calls Marble Coworkers. These are AI agents designed to execute the business processes using the software and tools that the company already owns. The goal is for the AI to replicate the actions of the human it observed, ensuring that the output remains consistent with the company's established standards.

A key feature of these agents is that they are not intended to replace human judgment entirely. The system is designed to allow for human involvement in specific areas that require high-level decision-making or final approvals. This creates a collaborative environment where the AI handles the repetitive, data-heavy steps while the human remains in control of the most complex parts of the workflow.

Privacy and data handling

Because the platform requires observing employee activity, Marble has implemented specific guardrails to manage privacy. The company states that all information gathered by the desktop application is anonymized before it is accessed by any human employees or AI models. Furthermore, the data is analyzed in aggregate rather than looking at the specific actions of a single individual.

The monitoring software is also designed to be non-intrusive regarding personal life. It does not monitor private applications, private browser tabs, or any activity that occurs outside of official working hours. These restrictions are intended to ensure that the tool remains focused strictly on business processes rather than personal data.

Employees also have a level of control over the monitoring process. They are able to pause the observation at any time during the day. By combining these privacy measures with aggregate data analysis, Marble attempts to provide a way for companies to capture operational knowledge without creating a surveillance environment for their staff.

Market traction and growth

Marble has already begun deploying its technology across a variety of industries and business types. The company reports that it has worked with organizations including EGYM, Mutares, and Enam. Its client base also includes various technology companies and businesses backed by private equity firms.

The startup recently secured $3.75 million in a pre-seed funding round to accelerate its growth. This round was co-led by Heartcore and Seedcamp, with participation from several high-profile angel investors. These investors include a board member of Datadog, a founder of Personio, and executives from McKinsey, Mistral AI, and Lovable.

The new capital is earmarked for several specific expansion goals. Marble plans to use the funds to scale its inference capabilities and grow its engineering team based in Munich. Additionally, the company intends to use the investment to support its upcoming launch into the United States market.

The future of automation

The work of Marble Cognition highlights a shift in how companies view the potential of AI. Rather than just looking for