What happened
Marble Cognition, a Munich-based AI startup founded in 2025, has closed a $3.75 million pre-seed funding round led by Heartcore Capital and Seedcamp. The company is developing a desktop application designed to observe employees as they interact with business software. By recording screen activity, mouse clicks, and keystrokes, the platform aims to generate structured process models that allow AI agents—branded as 'Coworkers'—to execute routine tasks autonomously. The company intends to use the capital to scale capabilities, expand its engineering team in Munich, and initiate a launch in the United States.
The funding round included participation from various angel investors with backgrounds in enterprise software and AI, including individuals associated with Datadog, Personio, McKinsey, Mistral AI, and Lovable.
Marble’s product thesis centers on the idea that AI agents can 'read' structured process maps derived from observation, allowing them to perform tasks within existing software stacks without requiring companies to migrate to new platforms.
The company claims its agents are designed to keep humans in the loop for tasks requiring judgment, focusing automation on the routine, repetitive paths of a workflow.
While the company has named EGYM, Mutares, and Enam as clients, the reports do not specify the nature of these engagements, such as whether they are paid contracts or pilot programs, nor do they detail the specific workflows being automated.
Why it matters
The core challenge in enterprise AI adoption is that most business processes are not formally documented, making it difficult to translate human workflows into machine-readable instructions. Marble Cognition’s approach attempts to bypass the need for manual standard operating procedures by deriving process maps directly from observed behavior. This method aims to capture the 'shortcuts' and practical workarounds that experienced employees use, which are often missing from official documentation. If successful, this could significantly lower the barrier to deploying AI agents in legacy environments that lack modern APIs, though it introduces significant privacy and security considerations regarding workplace monitoring.
Traditional automation methods, such as Robotic Process Automation (RPA) or process mining, have historically struggled with fragility when screen layouts change or when tasks involve complex human judgment between disparate systems.
By utilizing modern language models to interpret screen activity, Marble aims to create more resilient agents that can generalize across variations in a workflow, potentially making automation more accessible to non-technical teams.
The reliance on observing actual employee behavior rather than relying on static documentation addresses the 'documentation gap' that often stalls AI integration in large enterprises.
The company's focus on running inside existing tools is a strategic choice to reach legacy systems that lack clean integration points, though this exposes the agents to potential failures if the underlying software interfaces are updated.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
What to watch next
The primary uncertainty is whether the platform can maintain high accuracy in complex, real-world environments without requiring extensive human cleanup of the generated process maps. Potential buyers should monitor how the company handles sensitive data, such as password masking and keystroke privacy, particularly given the strict workplace monitoring regulations in Germany and the EU. Additionally, the company's ability to maintain agent stability when interface layouts change remains a technical hurdle. Observers should look for independent verification of the platform's performance beyond the initial pilot phase with named clients like EGYM, Mutares, and Enam.
The company has not disclosed its valuation, and claims regarding the speed of the funding round closure remain unconfirmed by independent sources.
Privacy and security remain the most significant barriers to adoption; the company has not yet provided public details on how it secures credentials or masks sensitive user data during the observation process.
The effectiveness of the 'Coworkers' agents will be tested as the company attempts to scale from pilot programs to broader enterprise deployments in the U.S. market.
The industry is currently evaluating whether agentic AI can deliver measurable ROI, and Marble will need to demonstrate that its process maps are accurate enough to reduce, rather than increase, the workload for human supervisors.