Zalos AI raised $3.6 million in seed funding to build AI agents that automate finance operations by directly manipulating ERP systems through their user interfaces. The round was led by 14 Peaks with participation from Cohen Circle, 20VC, and notable angels including FedEx CFO Mike Lenz and Happy Robot founder Pablo Palafox. The London-based startup is targeting the messy reality of enterprise finance teams who spend hours clicking through systems like SAP, Oracle, and QuickBooks.
This is part of a broader wave of "computer use" agents that interact with software like humans do—through screens, clicks, and forms—rather than requiring API integrations. It's a pragmatic approach to a real problem: most enterprise software wasn't built for AI integration, and retrofitting APIs is expensive and slow. Companies like Anthropic recently launched computer use capabilities for Claude, while others like Adept and MultiOn are building similar screen-based automation. The finance operations space is particularly ripe for this approach because it's process-heavy, rules-based, and often involves moving data between systems that don't talk to each other.
The challenge with these "human-like" AI agents is reliability. Finance operations require accuracy that current AI models struggle to guarantee, especially when navigating complex, dynamic interfaces. While the approach avoids integration headaches, it introduces new failure modes—what happens when the ERP interface changes, or when the AI misclicks on a critical transaction? Zalos will need to prove their agents can handle edge cases and maintain audit trails that satisfy finance teams and regulators.
For developers building similar automation tools, this validates the computer use approach but highlights the need for robust error handling and human oversight. The real test isn't whether these agents can automate happy-path scenarios, but whether they can fail gracefully when things go wrong.