Cognizant AI Lab

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

Next week, Babak Hodjat, Chief AI Officer at Cognizant, will be at Web Summit Lisbon speaking in two sessions on how AI is reshaping both product design and enterprise systems. 🔷 The Art of Building Intelligent Things 📅 Wednesday, Nov 12 | 16:00–16:20 | Centre Stage In this fireside, Babak joins Ciara Riordan to discuss how intelligent systems are shaped not just by data and models, but by product design, user experience, and emotional context. 🔷 The Agentic Future of Business: From Voice Assistants to Enterprise Intelligence 📅 Thursday, Nov 13 | 15:10–15:30 | AI Summit Track, Stage 4 In this keynote, Babak explores the evolution of AI agents and how they’re enabling collaboration, adaptability, and scale in enterprise environments. Don’t miss these two powerful sessions. #WebSummit #AI #Cognizant

Project Resilience is a collaborative platform that empowers a global community of innovators and thought leaders to apply AI and data for real-world impact. It was built under the Global Initiative on AI and Data Commons and supports real-world decision-making efforts that advance the UN Sustainable Development Goals. Anyone can make an impact with Project Resilience: 🔹 Decision makers: Use insights from data and models to inform impactful decisions 🔹 Technical builders: Gather data, train models, build UIs, and deploy apps that tackle challenges 🔹 General public: Explore projects and proposals, suggest policies, and share feedback 🔹 Moderators: Review and approve submitted proposals and projects and organize the community Looking to make an impact on AI for Good? We're looking for collaborators: https://cgnz.at/6041tGvsJ #ProjectResilience #SDG

As we enter a new era of artificial intelligence (AI) powered by multi-agent systems, these systems don’t just answer questions — they reason, coordinate, and refine their responses. However, to users, the deeper thinking that enables this intelligence can sometimes feel like a delay. In a World Economic Forum article, Cognizant's Chief AI Officer, Babak Hodjat and Global Head of Cognizant Moment, Benjamin Wiener explore how thoughtful UX can make the invisible turn visible and help users understand, trust, and collaborate with intelligent systems to effectively harness the next generation of AI. Read the full article here: https://cgnz.at/6049tDf5f #MultiAgentAI #Cognizant #Agents

The next generation of enterprise systems isn’t just connected, it’s intelligent. We’ve agentified our entire Cognizant intranet, turning it into an intelligent network of agents that can reason, act, and collaborate across functions. In the Intranet Agents With Coded Tools demo powered by our Cognizant AI Lab Neuro SAN, user requests are intelligently routed by an orchestrator agent to specialized sub-agents across departments like IT, Finance, Legal, and HR. Each sub-agent brings domain expertise and can trigger coded tools, such as APIs for leave balances or intranet updates. Explore the demo and get started on GitHub: https://cgnz.at/6001tD3RF #NeuroSAN #CognizantAILab

Our recent paper on ES Fine-Tuning has received an incredible response from the community. To make research in this space even more accessible, the research team has developed an accelerated implementation that achieves up to 10× faster total runtime. This advancement was made possible through a complete infrastructure refactor and integration with faster inference engines. As the researchers continue working on the next release with improved API design and greater usability, this milestone is being shared early in recognition of how valuable every GPU hour is for practitioners advancing AI innovation. Check out the accelerated version on Github: https://cgnz.at/6006t8g8e Read the full paper here: https://cgnz.at/6044t8brQ #LLMs #Finetuning #Cognizant #EvolutionStrategies