Maisa AI Secures $25M Funding to Revolutionize Enterprise AI with Accountable Agents

Maisa AI gets $25M to fix enterprise AI's 95% failure rate | TechCrunch

A staggering 95% of generative AI pilots at companies are failing, according to a recent report published by MIT's NANDA initiative. But rather than giving up on the technology altogether, the most advanced organizations are experimenting with agentic AI systems that can learn and be supervised. That's where Maisa AI comes in. The year-old startup has built its entire approach around the premise that enterprise automation requires accountable AI agents, not opaque black boxes. With a new, $25 million seed round led by European VC firm Creandum, it has now launched Maisa Studio, a model-agnostic self-serve platform that helps users deploy digital workers that can be trained with natural language. While that might sound familiar -- reminiscent of so-called vibe coding platforms like Cursor and the Creandum-backed Lovable -- Maisa argues that its approach is fundamentally different. "Instead of using AI to build the responses, we use AI to build the process that needs to be executed to get to the response -- what we call 'chain-of-work," Maisa CEO David Villalón told TechCrunch. The principal architect behind this process is Maisa's co-founder and Chief Scientific Officer, Manuel Romero, who had previously worked with Villalón at Spanish AI startup Clibrain. In 2024, the duo teamed up to build a solution to hallucinations after seeing firsthand that "you could not rely on AI," Villalón said. The pair isn't skeptical about AI, but they think it won't be feasible for humans to review "three months of work done in five minutes." To address this, Maisa employs a system called HALP, standing for Human-Augmented LLM Processing. This custom method works like students at the blackboard -- it asks users about their needs while the digital workers outline each step they will follow. The startup also developed the Knowledge Processing Unit (KPU), a deterministic system designed to limit hallucinations. While Maisa started out from this technical challenge rather than a use case, it soon found that its bet on trustworthiness and accountability resonated with companies hoping to apply AI to critical tasks. For instance, clients that currently use Maisa in production include a large bank, as well as companies in the car manufacturing and energy sectors. By serving these enterprise clients, Maisa hopes to position itself as a more advanced form of robotic process automation (RPA) that unlocks productivity gains without requiring companies to rely on rigid predefined rules or extensive manual programming. To meet their needs, the startup also offers them either deployment in its secure cloud or through on-premise deployment. This enterprise-first approach means Maisa's customer base is still very small compared to the millions flocking to freemium vibe-coding platforms. But as these platforms are now exploring how to win enterprise customers, Maisa is moving in the opposite direction with Maisa Studio, which is designed to grow its customer funnel and ease adoption. The startup also plans to expand with existing customers that have operations in multiple countries. With dual headquarters in Valencia and San Francisco, Maisa itself already has a foothold in the U.S., as reflected in its cap table; its $5 million pre-seed round last December was led by the San Francisco-based venture firms NFX and Village Global. In addition, TechCrunch learned exclusively that U.S. firm Forgepoint Capital International participated in this new round via its European joint venture with Spanish bank Banco Santander, highlighting its appeal for regulated sectors. Focusing on complex use cases demanding accountability from non-technical users could be a differentiator for Maisa, whose competitors include CrewAI and many other AI-powered, business-focused workflow automation products. In a LinkedIn post, Villalón highlighted this "AI framework gold rush," warning that the "quick start" becomes a long nightmare when you need reliability, auditability, or the ability to fix what went wrong." Doubling down on its goal to help AI scale, Maisa plans to use its funding to grow from 35 to as many as 65 people by the first quarter of 2026 in order to meet demand. Starting in the last quarter of this year, the startup anticipates rapid growth as it begins serving its waiting list. "We are going to show the market that there is a company that is delivering what has been promised, and that it's working," Villalón said.

Maisa AI reels in $25M for its AI agent development platform - SiliconANGLE

Maisa AI Inc., a startup with a platform for building artificial intelligence agents, today announced that it has raised $25 million in seed funding. Early Spotify Technology SA backer Creandum led the investment. It was joined by Forgepoint Capital International, NFX and Village Global. The latter two funds invested $5 million in Maisa AI last December. Maisa AI's platform enables business users to create AI agents using natural language prompts. According to the company, its software doesn't require programming expertise or training datasets to use. Agents created with the platform can automate tasks such as extracting data from forms and scanning network traffic for malicious activity. "Users can scale AI at pace, do so safely and without the need for an entire development team to support," said Maisa AI co-founder and Chief Executive Officer David Villalon. The company says that its AI agents can generate a step-by-step explanation of how they complete each task. That allows users to verify the work was carried out correctly. If an agent makes a mistake, workers can provide natural language feedback explaining what should be improved. Maisa AI's platform can resolve some processing obstacles automatically. The software includes a so-called self-sealing mechanism that allows I agents to detect when task requirements change and adapt accordingly. Additionally, agents can ask for additional instructions or access to a third-party application if they can't complete a task based solely on the user's initial prompt. Under the hood, Maisa AI's platform is powered by a software engine called KPU. When a worker asks an agent to perform a task, KPU plans how the task should be carried out using a large language model. It then sends the plan to a component that completes the task and generates feedback on how it could be carried out more efficiently in the future. The company first detailed KPU last March. It debuted a new release, Vinci KPU, in November. The upgraded version provides lower latency and supports test-time compute, which means it can increase the amount of infrastructure used during prompt processing to increase output quality. Vinci KPU also includes what Maisa AI describes as an "integrated knowledge base consultation system." That mechanism allows the software to verify the accuracy of AI agents' prompt responses. Maisa AI says that Vinci KPU can find the information necessary to complete tasks more cost-efficiently than RAG, or retrieval-augmented generation, tools.