Cognizant AI Lab

Long-horizon reliability has long been one of the toughest challenges for LLMs. Apple’s Illusion of Thinking study underscored this fragility, highlighting how quickly models lose accuracy as reasoning chains grow. Last week, we announced our AI Lab’s new approach that charts a path forward: a decomposed, micro-agent–based system capable of completing over one million dependent reasoning steps with zero errors. This video by ReInikeAI breaks down our paper and walks through the key insights. Watch the full video: https://cgnz.at/6040tnppY Read our full paper: https://cgnz.at/6041tnppl #AIAgents #AI #AIResearch

At the recent Web Summit in Lisbon, Cognizant’s Chief AI Officer, Babak Hodjat joined Sky News to highlight an important message: we must avoid over-trusting AI and stay grounded in understanding both its capabilities and its limitations. Read the full interview here: https://cgnz.at/6049tTYLS #AI #AgenticAI

We’re excited to spotlight a new MIT Press book titled Neuroevolution: Harnessing Creativity in AI Agent Design written by long-time contributors to the field Sebastian Risi, Yujin Tang, David Ha, and Risto Miikkulainen. The book brings together several decades of research that have transformed evolutionary computation from an academic pursuit into a foundation for building adaptive and creative AI. Read it for free at: https://cgnz.at/6048traUM Learn more with the blog: https://cgnz.at/6049traU3 Watch the video here: https://cgnz.at/6040traUO The website also comes complete with demos, tutorials, exercises, and teaching resources to allow you to get hands-on with neuroevolution and creative agent design. #Neuroevolution #AIAgent #AI

How can enterprises balance innovation, productivity, and ethics in the era of AI? In the latest episode of The Agent Effect, Leo S Mackay, Jr., Senior Vice President of Ethics and Enterprise Assurance at Lockheed Martin and Cognizant board member, joins host Paul Jarratt to discuss how AI and multi-agent systems are redefining enterprise performance and what it takes to adopt them responsibly. From the power of trusted partnerships to the importance of ethical foundations, this conversation offers an inside look at how business and technology leaders can prepare for the agentic future. Spotify: https://cgnz.at/6044tR8Kq Apple Podcast: https://cgnz.at/6045tR8KS #TheAgentEffect #AIEthics #AIAgents

On Thursday, November 13th, Cognizant's Chief AI Officer Babak Hodjat spoke at Web Summit Lisbon in a session titled “The Agentic Future of Business: From Voice Assistants to Enterprise Intelligence.” He outlined AI’s progression from early monolithic systems to the small specialized agents of the 80s and 90s, leading to today’s multi-agent architectures built around powerful LLMs. Babak also announced our newly published research paper, “Solving a Million-Step LLM Task with Zero Errors" which responds to Apple’s finding from the "Illusion of Thinking" study that chain-of-thought models collapse after a few hundred reasoning steps. Our paper shows that reasoning can be distributed across thousands to a million agents, enabling perfectly accurate long-horizon reasoning. Using this approach, the team solved a 20-disk Tower of Hanoi with one million coordinated agents and zero errors, demonstrating that large-scale multi-agent systems can achieve levels of reliability and reasoning depth that single LLMs cannot, pointing toward a future where businesses and applications operate as interconnected agent webs. Read the full blog here: https://cgnz.at/6040tP3sn #WebSummit #Cognizant #AIAgents #MultiAgentSystems

Enterprises are turning to AI to automate long, complex processes — from supply chain planning to claims handling and clinical operations. But today’s large language models struggle with multi-step reasoning. Small mistakes compound, and in enterprise workflows, small mistakes aren’t small. Apple’s Illusion of Thinking study showed this clearly: even the most advanced LLMs break down on long, complex reasoning tasks as tiny errors pile up. Today, we’re announcing new research demonstrating the first AI system to complete more than one million dependent reasoning steps with zero errors. Our approach, called MAKER, uses extreme task decomposition and microagent voting to sustain accuracy far beyond what a single LLM can achieve. This work suggests the future of AI isn’t just bigger models, but smarter, structured systems designed for reliability. Read the blog: https://cgnz.at/6049t32mN Read the full paper: https://cgnz.at/6040t32m4 #AgenticAI #LLMs

AI needs to understand how humans are doing work—not the other way around. Context gives AI data to understand how teams are getting their work done to provide more relevant and valuable information. Babak Hodjat and Rohan Murty share what this looks like with Neuro AI at #CognizantDiscovery2025.

Neuroevolution — the fusion of evolutionary algorithms and neural networks — is redefining how we explore artificial and biological intelligence. By mimicking natural selection instead of relying solely on gradient descent, it evolves neural architectures that can learn, adapt, and even surprise. As Risto Miikkulainen notes, this approach helps reveal why certain brain circuits exist, how they develop, and how intelligent behavior may have emerged. By linking evolution, learning, and neuromodulation, neuroevolution offers a fresh lens on cognition and creativity. Read more here: https://cgnz.at/6047tJpxJ #AI #Neuroevolution #ArtificialIntelligence