Marble Cognition has secured $3.75M in pre-seed funding to build what it calls artificial intelligence (AI) coworkers, software that learns a company’s processes by observing employees at work. Venture capital firms Heartcore and Seedcamp co-led the round for the Munich startup, which was founded late last year. Angel investors also joined, including a Datadog board member, a founder of HR software company Personio and executives from McKinsey, Mistral AI and Lovable.
The money will go towards a larger engineering team in Munich, a launch in the United States and more inference capacity, meaning the computing power needed to run its AI models.
Co-founder Felix Rösner summed up the problem the startup wants to solve. “Every company we talk to wants AI to take over real work. But nobody can tell the AI how that work is actually done,” he said.
How The App Works
Marble’s desktop app records screen activity, clicks and keystrokes across the business software employees use and converts that activity into process maps, step-by-step records of how each task is done. From the maps, the startup builds AI agents called Marble Coworkers that carry out the same tasks inside a company’s existing tools, while people stay involved in steps that need judgement or approval.
Management consultancy McKinsey estimated last year that existing technology could in theory automate about 57% of working hours in the United States. Marble points to a practical hurdle in its own client base, saying around 90% of core business processes at those companies have never been formally written down.
The startup’s website says a few weeks are enough to show how a team works and which tasks to hand over first. The uses it lists range from logging claims and preparing quotes at insurers to matching payments at financial services firms and running payroll and closing the monthly books at professional services firms.
Larger software companies are pursuing a similar goal by a different route. ServiceNow released its AI Workflow Factory in October. The product analyses the event logs that business systems keep, a method known as process mining, to find workflows that need fixing and then uses AI agents to rebuild them. Marble’s method starts from what people do on screen instead.
Privacy On The Desktop
Recording employees’ screens raises workplace privacy questions, and Marble has built several limits into its product. The company says it anonymises the data and studies activity in aggregate rather than person by person. The app ignores private apps, private browser tabs and anything outside working hours, and employees can pause recording at any time.
Its website adds that personal data is removed before processing and that the full anonymisation step can run inside a customer’s own systems. The startup has completed a System and Organization Controls (SOC) 2 Type I audit, an independent review of how its security controls are designed. The company also lists compliance with the General Data Protection Regulation, the European Union’s data protection law, and offers hosting in the bloc.
Founders And Early Customers
Rösner met co-founder Christophe Schmit while both were studying at the Center for Digital Technology and Management in Munich. He previously worked on digital workers, AI software that handles routine office tasks, at 11x, a startup backed by Benchmark and Andreessen Horowitz. Schmit came from Enam, a company backed by investor Elad Gil that buys businesses and uses AI agents to run their operations.
Enam is now among Marble’s customers, alongside Munich fitness technology company EGYM and listed investment group Mutares. The startup says it also works with other technology firms and with businesses owned by private equity investors. Its website lists private equity among its target industries, with services for running a fund’s portfolio companies and combining businesses after a merger.
Between them, Marble’s founders bring experience from two companies built on AI agents. Their startup begins on the employee’s screen rather than in the company’s systems, mapping how work is done before an agent takes over any task.
