A shocking data is causing a stir in the tech circle: The latest report from MIT's NANDA project shows that 95% of enterprise generative AI pilot projects have failed. However, while most companies are disillusioned with AI technology, some cutting-edge organizations are turning to a new solution - learnable and supervised intelligent agent AI systems.
Under this context, the startup Maisa AI, which was established only a year ago, has stood out with its unique corporate automation concept. The company firmly believes that enterprise-level automation requires accountable AI agents rather than opaque black-box systems. After completing a $25 million seed round led by the renowned European venture capital firm Creandum, Maisa AI officially launched Maisa Studio - a model-agnostic self-service platform that helps users deploy digital employees trained through natural language.
Disrupting the Traditional: From Generating Responses to Building Processes
Although this may sound similar to platforms like "atmospheric programming" such as Cursor and Lovable supported by Creandum, Maisa AI emphasizes a fundamental difference in its approach. "We don't use AI to build responses, but to build the processes needed to obtain those responses - we call it 'workchain'," said David Villalón, CEO of Maisa AI, in an interview with TechCrunch.
The main architect of this innovative process is Manuel Romero, co-founder and Chief Science Officer of Maisa, who previously worked with Villalón at the Spanish AI startup Clibrain. In 2024, the two collaborated on a solution to address AI hallucination issues, as they had witnessed the reality that "you can't rely on AI."
HALP System: Working Like a Student in Front of a Blackboard
These founders are not skeptical of AI, but they believe it's unrealistic to expect human review of three months' worth of work completed in five minutes. To solve this problem, Maisa adopted a system called HALP, or Human-Augmented Large Language Model Processing System. This customized approach works like a student demonstrating on a blackboard - the digital employee asks the user for their needs and explains each step they will follow in detail.
The company also developed a Knowledge Processing Unit (KPU), a deterministic system designed to limit hallucination phenomena. Although Maisa approached the issue from a technical challenge rather than use cases, they soon found that their investment in trustworthiness and accountability resonated strongly with enterprises looking to apply AI to critical tasks. Current clients using Maisa in production include a major bank, as well as companies in the automotive manufacturing and energy sectors.
Enterprise Deployment: Security in the Cloud and On-Premise
Through serving these enterprise customers, Maisa hopes to position itself as a more advanced form of robotic process automation (RPA), unleashing productivity potential without requiring enterprises to rely on rigid pre-defined rules or extensive manual programming. To meet customer needs, the company also offers options for secure cloud deployment or on-premise deployment.
This enterprise-first strategy means that Maisa's customer base is still small compared to the millions flocking to free-tier atmospheric programming platforms. But as these platforms begin to explore how to win over enterprise customers, Maisa is taking the opposite approach, expanding its customer funnel and simplifying adoption through Maisa Studio.
AIbase learned that the company also plans to expand with existing customers that operate in multiple countries. Maisa has dual headquarters in Valencia and San Francisco, and has established itself in the U.S., which is reflected in its equity structure - the $5 million pre-seed round in December was led by the San Francisco-based venture capital firm NFX and Village Global.
Additionally, TechCrunch独家获悉, the U.S. company Forgepoint Capital International participated in this funding round through its European joint venture with Santander Bank in Spain, highlighting its appeal in the regulated industry.
Differentiated Competition: Focused on Complex Scenarios and Non-Technical Users
Focusing on complex use cases that require accountability from non-technical users could be Maisa's differentiating advantage, as its competitors include CrewAI and many other AI-driven business workflow automation products. In a LinkedIn post, Villalón emphasized the "AI framework gold rush," warning that when you need reliability, auditability, or the ability to fix errors, "quick start" can become a long-term nightmare.
To help scale AI, Maisa plans to use the funds to expand its team from 35 to 65 people, expecting to complete the expansion by the first quarter of 2026 to meet demand. Starting from the last quarter of this year, the company expects rapid growth and will begin serving customers on its waiting list. Villalón said, "We will show the market that there is a company fulfilling its promises, and it really works."
Facing the harsh reality of a 95% failure rate for enterprise AI projects, Maisa AI's accountable intelligent agent solution provides a new path for enterprise AI applications. This technology route emphasizing transparency and supervisory capabilities may become the new benchmark for enterprise AI deployment.