Cognizant AI Lab Wrapped 2025

A year of building agentic, real-world AI at scale.

2025 was a year of building at Cognizant AI Lab.

We moved beyond asking what is possible and focused on building AI that works in the real world. Our work made AI more agentic, more adaptive, and more aligned with how decisions actually happen.

Throughout the year, we launched tools, published new research, and brought our work into global forums and developer communities. From multi-agent systems to neuroevolution, everything we built pushed toward one goal: smarter decision-making at scale.

Let’s rewind 2025.

2025: By the Numbers

This year, we:

• Published 18 peer-reviewed research papers

• Open sourced 2 major tools: the Neuro AI Multi-Agent Accelerator and Evolution Strategies Toolkit

• Were granted 3 AI patents 

• Presented at 25+ global research, industry, and policy forums

• Released the MIT book Neuroevolution: Harnessing Creativity in AI Agent Design

• Published first 8 episodes of our new podcast The Agent Effect

• Were featured in 7 broadcast segments on AI agents and scalable decision systems

• Earned the HUMIES Gold Award at GECCO 2025 for real-world evolutionary AI

• Built 4 AI for Good projects through Project Resilience

• Expanded our global presence with the launch of Cognizant AI Lab India

Agents Took Center Stage

Agentic AI was the most consistent thread across everything we built in 2025.

This year, we moved beyond static models and focused on systems that can plan, act, adapt, and collaborate in real decision environments. Our work on agentic and multi-agent systems shaped tools, research, and public conversations throughout the year.

Top moments

• Shattering the Illusion: MAKER Achieves Million-Step, Zero-Error LLM Reasoning: Our new approach shows how breaking reasoning across millions of AI agents can achieve unprecedented reliability, pointing to a practical path for scaling LLM intelligence to organizational and societal levels

• How Neuro AI Helped Make History at the Cognizant Vibe Coding Event: We used the Neuro® AI Multi-Agent Accelerator to build a multi-agent evaluation network that reviewed 30,000+ hackathon submissions in under 24 hours, supporting the judging process for Cognizant’s Guinness World Record–breaking event. 

• Neuro SAN is All You Need: Our open-sourced Neuro® AI Multi-Agent Accelerator outperforms leading multi-agent frameworks across areas such as communication, interoperability, secure data, and more.

Research We Published

The research published by our researchers truly breaks barriers of agentic AI. Here are a few that have redefined AI research:

• Meyerson, E., Paolo, G., Dailey, R., Shahrzad, H., Francon, O., Hayes, C. F., Qiu, X., Hodjat, B., and Miikkulainen, R. (2025) Solving a Million-Step LLM Task with Zero Errors: Building on insights from Apple’s Illusion of Thinking paper, which showed that LLMs fail on long dependency chains, this work introduces our approach with MAKER, the first system to solve a million-step reasoning task by decomposing a 20-disk Towers of Hanoi problem into validated microagent subproblems, achieving zero errors.

• Qiu, X., Gan, Y., Hayes, C. F., Liang, Q., Meyerson, E., Hodjat, B., and Miikkulainen, R.(2025) Evolution Strategies at Scale: LLM Fine-Tuning Beyond Reinforcement Learning: We introduced the first successful use of evolution strategies (ES) to fine-tune LLMs with billions of parameters, marking a transformative new approach beyond traditional reinforcement learning (RL) methods.

• Miikkulainen, R. (2025). Neuroevolution Insights Into Biological Neural Computation. Science 387, eadp 7478. In neuroevolution, a population of neural network encodings is evolved based on how well each network’s behavior solves a task. This review article discusses how neuroevolution experiments can provide insight into the evolutionary origins of biological neural circuits, behavior, and cognitive processes.

• Young, D., Francon, O., Meyerson, E., Schwingshackl, C., Bieker, J., Cunha, H.,Hodjat, B., and Miikkulainen, R. (2025). Discovering Effective Policies for Land-Use Planning. With historical land-use data and a carbon budget model, we developed a NeuroAI system  to decide where and how land use could be changed optimally.

• Meyerson, E., & Qiu, X. (2025). Position: Scaling LLM Agents Requires Asymptotic Analysis with LLM Primitives. ICML. This paper argues that scaling LLM agent systems requires formal asymptotic analysis, treating the LLM forward pass as the basic unit of computation to reason about and improve multi-agent system efficiency.

In the Headlines 

This year, our work reached broader audiences through media coverage and press features.

These moments helped bring agentic AI and applied research into wider conversations beyond the lab and developer community.

Top moments

• Cognizant opens AI Lab and Moment Studio in Bengaluru to push enterprise AI: As part of its $1 billion AI commitment, Cognizant strengthened its global footprint by launching an AI Lab and Moment Studio in Bengaluru.

• 'Agentic AI' is current industry buzzword - but what does it mean and should we be cautious?: 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, stay grounded in understanding both its capabilities and its limitations and use it responsibly. 

• CNBC: Cognizant's New AI Lab: Inside The Silicon Valley: CNBC-TV18 visited Cognizant's New AI Lab in San Francisco. In this video, we did a deep dive into Cognizant's billion dollar AI bet, the role of humans in an AI era and Cognizant's mega deals. 

Built for Developers and Partners

A major focus of 2025 was making our work usable. We put tools, frameworks, and ideas directly into the hands of developers, partners, and communities so they could build, test, and extend agentic systems themselves.

Top moments

• Open-sourced the Neuro® AI Multi-Agent Accelerator, giving developers and researchers a practical way to prototype, customize, and deploy scalable agent networks for real-world decision-making

• Released open-source evolution strategies fine-tuning code, giving developers a scalable, GPU-efficient way to fine-tune large language models beyond reinforcement learning

AI for Good in Action

In 2025, we continued building AI systems designed to address real-world challenges through Project Resilience.

Project Resilience is a collaborative platform that empowers a global community of innovators and thought leaders to apply AI and data to advance the UN Sustainable Development Goals. Together with partners and contributors, we translated this work into concrete projects that support real decision-making across sustainability, resilience, and global impact efforts.

Top projects

• Water System Chlorination Optimization:
  We developed an agentic control system to optimize real-time chlorination in water distribution networks, improving safety while minimizing chemical use. This project was presented at IJCAI 2025.
  
  
• Land Use Optimization:
  We applied evolutionary AI to help optimize land-use decisions by trading off carbon emissions and economic impact, supporting climate-aware planning for governments and landowners. Read the blog for a breakdown on the research behind the project.
  
  
• Irrigation Strategy Optimization:
  We built an AI-driven decision-support system using the AquaCrop simulator to help farmers and policymakers optimize irrigation strategies, balancing crop yield, water use, and field management costs. This work was presented at the FAO workshop at the AI for Good Summit 2025.

Thought Leadership + The Agent Effect

Beyond publications and platforms, Cognizant AI Lab played a central role in shaping global dialogue on AI’s future. Lab leaders spoke at premier forums including WEF Davos, ICML, NeurIPS, GECCO, Web Summit, Microsoft Ignite, and major policy venues such as the OECD and GPAI Summit.

We also introduced The Agent Effect, a new podcast that brings together AI pioneers, technology leaders, and business executives to explore how agentic AI can drive enterprise transformation. Through conversations with leaders such as Kim Krogh Andersen, Product & Technology Group Executive at Telstra, Leo Mackay, SVP, Ethics and Enterprise Assurance,  and our own CEO, Ravi Kumar S, we break down what agentic AI really means and how its applications are reshaping enterprises.

Wrapping Up 2025

2025 marked clear growth for Cognizant AI Lab, where we demonstrated how agentic and neuro-evolutionary AI can operate reliably at real-world scale. Our work advanced long-horizon reasoning, scalable learning methods, and multi-agent collaboration, while remaining grounded in responsible deployment and human-centered design.

As we move into 2026, our focus remains unchanged: to build adaptive, trustworthy AI systems that align with how decisions are made in the real world. And to do so with rigor, transparency, and impact.

Thank you to our collaborators, partners, and community for shaping this journey. We look forward to continuing the work ahead.