Evos raises $9M pre-seed to build AI agents for industrial tasks

Evos’s agents—software tools, not robots—handle tasks like reconciling invoices, managing freight loads, and chasing missing documents. They learn from human decisions and gradually take on more responsibility as they prove reliable. The system starts with human approval gates, where employees review and refine AI suggestions before the agents operate independently. Evos reports 97% accuracy in auditing less-than-truckload carrier invoices for one client, processing nearly 3,500 decisions and $600,000 in invoice value over two months. However, these figures are use-case specific and not platform-wide guarantees.

The company’s architecture combines real-time data from business systems, a controller managing over 300 specialized subagents, and customer-specific small language models. Success hinges on aligning data freshness, decision quality, permissions, and escalation protocols. Early results suggest potential for reducing manual workloads, but human oversight remains critical for exceptions and high-stakes decisions.