Paris-based startup Twin Labs is pioneering the development of an automation product designed for streamlining repetitive tasks, ranging from onboarding new employees across internal services to managing stock levels and generating financial reports from various SaaS products. CEO Hugo Mercier explained that the inspiration behind Twin Labs was a science-fiction concept exploring the duplication of human actions through training an AI agent. Notably, the startup employs multimodal models, particularly GPT-4 with Vision (GPT-4V), to mimic human tasks, distinguishing itself from conventional methods.

Initially experimenting with traditional Language Model Models (LLMs), Twin Labs found them unreliable in decision-making. Transitioning to GPT-4V, trained extensively on diverse software interfaces, allowed the company to understand features behind buttons, enhancing task performance. In contrast to competitors like Zapier, Twin Labs operates more like a web browser, automating processes without relying on complex APIs and multi-step procedures. While the AI assistant is still in development, the startup raised $3 million in pre-seed funding from various investors, including Betaworks, Motier Ventures, and angel investors like Florian Douetteau and Romain Huet.

Twin Labs aims to tackle challenges in its autonomous agent system, acknowledging the current high costs of task completion. The startup plans to initially release a product with a library of pre-trained tasks before opening its platform for clients to create customized tasks. In a departure from common AI product interfaces, Twin Labs focuses on a hands-on approach, aiming to simplify daily tasks and alleviate user burdens. The company's innovative vision is shaping a unique interaction model with AI models, promising a significant shift in the automation landscape.