The figures accumulate and translate the same reality, the adoption of artificial intelligence in business is struggling to go beyond the experimental stage. According to MIT Nanda Initiative, 95 % of generative AI pilots fail and bring no measurable impact to the income statement. IDC estimates that 88 % of projects never take the production stage. For its part, the BCG observes that only 22 % of companies have managed to draw a concrete value from AI beyond the proof of concept.
In this context, buta is based on a reliability and traceability approach. The startup, founded in 2024 in Valence and San Francisco, has developed a owner engine, the Knowledge Processing Unit (KPU). This transforms generalist language models into specialized agents capable of performing tasks according to an explicit computational logic. Each step is recorded in a “chain-of-work”, an auditable register that documents all of the reasoning followed.
The company presents this method as a response to the recurring problem of hallucinations. “Our platform gives teams the power to use AI workers who not only perform with great precision and intelligence, but who explain and prove their logic and document each stage of the work process,” explains David Villalon, co -founder and CEO.
The company has just launched Maisa Studio, a solution allowing business experts, without technical training, to create agents in natural language. These “Digital Workers” are based on a method called HALP (Human-Augmented LLM Processing) which allows them to learn in a situation, without requiring prior datasets or specialized engineers. Deployment can be done in the secure Cloud of COUSA or in private installation to meet compliance constraints.
The first use cases concern players in the bank, energy and automobile. An investment bank has replaced its manual media monitoring process with MOIA agents capable of extracting relevant facts, assessing reputational risk and generating auditable summaries in minutes. A financial service company has automated its transaction controls, eliminating 99 % of false positives and multiplying the productivity of its teams by ten.
Maisa has just completed a table of 21.4 million euros in Seed, led by Creandum, with the participation of ForgePoint Capital (via his joint venture with Banco Santander), NFX and Global Village. This operation comes eight months after a pre-seed of 4.7 million euros gathered with NFX and Village Global in December 2024. The company, co-founded by David Villalon, former Chief Ai Officer de Clibrain, and Manuel Romero, major contributor to Huggingface, plans to use this funding to strengthen its R&D, recruit in engineering, sales and customer relationship teams, expansion in Europe and North America.
