Founded only at the end of 2025, Munich AI startup Marble Cognition has now secured around €3.35 million ($3.75 million) in a pre-seed funding round led by Heartcore and Seedcamp. The company plans to use the capital to further develop its AI agents, which learn workflows by observing them and then automate them.
Instead of laboriously documenting existing processes, Marble has its AI first observe how employees actually work. From this, the startup creates structured process models, which serve as the basis for so-called “Marble Coworkers” to take over recurring tasks. Its customers already include companies such as Egym, Mutares, and Enam.
When companies want to deploy artificial intelligence, they repeatedly face the key challenge that many workflows in enterprise environments are not fully documented. According to Marble, nearly all previous clients have around 90 percent of their core processes stored only in emails, spreadsheets, or the heads of individual employees. The startup aims to close precisely this information gap.
Heartcore and Seedcamp Lead Pre-Seed Round
The funding round is co-led by venture capital firms Heartcore and Seedcamp. In addition, business angels from the international tech and consulting scene are participating, including a Datadog board member, a Personio co-founder, and executives from McKinsey, Mistral AI, and Lovable.
The investors see capturing and structuring internal company processes in particular as an essential building block for the next stage of AI automation.
Reshma Sohoni, General Partner at Seedcamp, points to the challenge of making existing workflows transparent enough for automation in the first place. Marble provides companies with the necessary context for this.
Max Niederhofer, Partner at Heartcore, also sees the critical bottleneck not in the AI models themselves, but in accurately capturing actual work processes. Marble can derive standardized process descriptions from this and deploy AI agents to execute them.
With fresh capital, Marble Cognition plans to expand into the US market
The 3.35 million euros (3.75 million dollars) will primarily go toward the technical expansion of the platform, growing the engineering team in Munich, and entering the US market.
Marble Cognition is thus positioning itself in a growing market for agent-based AI systems. While numerous providers simplify the development and deployment of AI agents, the Munich startup focuses on the upstream question of which processes can actually be automated and how they really work in day-to-day business operations.
Combining process analysis with subsequent automation could be particularly relevant for companies whose workflows are spread across multiple applications and have so far relied heavily on experiential knowledge.
Crucial for its future development will be whether Marble can scale the approach beyond individual client projects. This also includes questions of data security, employee acceptance, and transferability to different industries.
Marble Coworker: AI Agents Learn Workflows by Observing
Marble’s technology starts even before actual automation takes place. A desktop application captures how employees use enterprise software, which steps they carry out, and how different applications interact. This creates a structured representation of the processes, which AI agents can then use to perform tasks.
The so-called Marble Coworkers then take over workflows within the existing systems. Employees remain involved for decisions that require human judgment.
According to the company, the recorded information is anonymized before humans or AI models can evaluate it. Private applications, private browser tabs, and activities outside of working hours are not recorded. Employees can pause tracking at any time. Marble states that the analysis is performed at an aggregate level rather than an individual employee level.
90 Percent Shorter Processing Time
Marble provides an initial use case from the financial services sector. At a US company owned by a private equity firm, a Marble Coworker was able to reduce processing time per case by 90 percent, according to company reports.
In the very first month, the team handled 50 percent more cases without hiring additional staff. The AI agent prepares settlement offers, which are then reviewed and approved by a specialist.
The underlying process was not automated using existing documentation. Instead, Marble Cognition learned the individual work steps by observing the team.
