Maisa Ai Gets $25m To Fix Enterprise Ai’s 95% Failure Rate

A staggering 95% of generative AI pilots astatine companies are failing, according to a caller report published by MIT’s NANDA initiative. But alternatively than giving up connected nan exertion altogether, nan astir precocious organizations are experimenting pinch agentic AI systems that tin study and beryllium supervised.

That’s wherever Maisa AI comes in. The year-old startup has built its full attack astir nan premise that endeavor automation requires accountable AI agents, not opaque achromatic boxes. With a new, $25 cardinal seed information led by European VC patient Creandum, it has now launched Maisa Studio, a model-agnostic self-serve level that helps users deploy integer workers that tin beryllium trained pinch earthy language.

While that mightiness sound acquainted — reminiscent of alleged vibe coding platforms for illustration Cursor and nan Creandum-backed Lovable — Maisa argues that its attack is fundamentally different. “Instead of utilizing AI to build nan responses, we usage AI to build nan process that needs to beryllium executed to get to nan consequence — what we telephone ‘chain-of-work,” Maisa CEO David Villalón told TechCrunch.

The main designer down this process is Maisa’s co-founder and Chief Scientific Officer, Manuel Romero, who had antecedently worked pinch Villalón astatine Spanish AI startup Clibrain. In 2024, nan duo teamed up to build a solution to hallucinations aft seeing firsthand that “you could not trust connected AI,” Villalón said.

The brace isn’t skeptical astir AI, but they deliberation it won’t beryllium feasible for humans to reappraisal “three months of activity done successful 5 minutes.” To reside this, Maisa employs a strategy called HALP, opinionated for Human-Augmented LLM Processing. This civilization method useful for illustration students astatine nan blackboard — it asks users astir their needs while nan integer workers outline each measurement they will follow.

Image Credits:Maisa AI

The startup besides developed nan Knowledge Processing Unit (KPU), a deterministic strategy designed to limit hallucinations. While Maisa started retired from this method situation alternatively than a usage case, it soon recovered that its stake connected trustworthiness and accountability resonated pinch companies hoping to use AI to captious tasks. For instance, clients that presently usage Maisa successful accumulation see a ample bank, arsenic good arsenic companies successful nan car manufacturing and power sectors.

By serving these endeavor clients, Maisa hopes to position itself arsenic a much precocious shape of robotic process automation (RPA) that unlocks productivity gains without requiring companies to trust connected rigid predefined rules aliases extended manual programming. To meet their needs, nan startup besides offers them either deployment successful its unafraid unreality aliases done on-premise deployment. 

This enterprise-first attack intends Maisa’s customer guidelines is still very mini compared to nan millions flocking to freemium vibe-coding platforms. But arsenic these platforms are now exploring really to triumph endeavor customers, Maisa is moving successful nan other guidance pinch Maisa Studio, which is designed to turn its customer chimney and easiness adoption.

The startup besides plans to grow pinch existing customers that person operations successful aggregate countries. With dual office successful Valencia and San Francisco, Maisa itself already has a foothold successful nan U.S., arsenic reflected successful its headdress table; its $5 cardinal pre-seed round past December was led by nan San Francisco-based task firms NFX and Village Global. 

In addition, TechCrunch learned exclusively that U.S. patient Forgepoint Capital International participated successful this caller information via its European associated venture pinch Spanish slope Banco Santander, highlighting its entreaty for regulated sectors.

Focusing connected analyzable usage cases demanding accountability from non-technical users could beryllium a differentiator for Maisa, whose competitors see CrewAI and galore different AI-powered, business-focused workflow automation products. In a LinkedIn post, Villalón highlighted this “AI model golden rush,” informing that nan “quick start” becomes a agelong nightmare erstwhile you request reliability, auditability, aliases nan expertise to hole what went wrong.”

Doubling down connected its extremity to thief AI scale, Maisa plans to usage its backing to turn from 35 to arsenic galore arsenic 65 group by nan first 4th of 2026 successful bid to meet demand. Starting successful nan past 4th of this year, nan startup anticipates accelerated maturation arsenic it originates serving its waiting list. “We are going to show nan marketplace that location is simply a institution that is delivering what has been promised, and that it’s working,” Villalón said.