Why Multi-Agent AI Outpaces Monolithic Models: Cognizant’s AI Outlook

The world of enterprise AI is undergoing a quiet but powerful transformation. Not just in how capable AI is—but in how it’s built. For years, businesses leaned on giant, monolithic models hidden behind opaque APIs. They were powerful, yes, but rigid, expensive, and often a black box. Now, the tide is shifting toward smaller, specialized AI agents that don’t try to do everything, but instead work together like a well-orchestrated team.

For Babak Hodjat, CTO of AI at Cognizant, this shift isn’t just technical—it’s healthy, even overdue.

Babak has seen AI hype come and go. He was building AI systems back in the 1980s, long before ChatGPT became a household name, and he holds patents that helped power Apple’s Siri. He’s not chasing flashy demos or viral buzz. His obsession is with AI that actually makes businesses stronger, faster, and more resilient in the real world.

When speaking with Techopedia, he laid out why this matters and where the future is headed.

Why Multi-Agent AI Outpaces Monolithic Models: Cognizant’s AI Outlook

Key Takeaways:

Agents act. Models just respond. A model can generate text or predictions, but an agent takes that output and does something meaningful with it—deciding, executing, and collaborating.

Open-source AI isn’t just cheaper—it’s empowering. Enterprises want control, transparency, and adaptability, not vendor lock-in. Open ecosystems make AI less of a mystery and more of a tool you can trust.

Trust is about actions, not words. Governance in AI won’t just be about what the model says—it’ll be about what the code actually does. Businesses need accountability at the level of behavior, not just output.

Multi-agent systems are the future of enterprise workflows. Imagine not one AI assistant, but a whole team of specialized agents handling tasks, coordinating, and solving problems together—like a digital workforce.

Agents won’t just serve humans—they’ll negotiate with each other. The next era of AI won’t be limited to “human asks, AI answers.” Instead, agents will interact directly, brokering deals, aligning goals, and optimizing processes across organizations.

Why DeepSeek Wasn’t a Sputnik Moment

Do you remember that buzz earlier this year when DeepSeek’s models hit the scene? For a moment, the tech world felt a mix of thrill and unease. Social feeds were flooded with headlines: small models, open-source, surprisingly capable. It felt like a shift. But Babak urged us to keep perspective.

“The reason why I thought it wasn’t a Sputnik moment is because it wasn’t a rethinking of how we do AI,” he said. “It was just an optimization.”

That wasn’t meant as criticism. In fact, there was a kind of respect in his tone. What DeepSeek showed was not a revolution, but a refinement—a signal that we’ve entered a new chapter where efficiency begins to matter more than brute computational power.

Think about it: instead of throwing bigger GPUs at the problem, they asked smarter questions about training and architecture. The result? Smaller models, less compute hunger, and surprisingly strong performance. Suddenly, you don’t need a massive server farm to run sophisticated agents. A high-end laptop is enough.

That’s not science fiction anymore—it’s here.

AI for Everyone, Not Just the Giants

One of the most powerful takeaways from Babak’s reflections is this: AI is finally beginning to decentralize. DeepSeek didn’t invent a new paradigm, but it did prove what’s possible when you mix openness with careful engineering.

“I can download a 14 billion parameter DeepSeek model onto my laptop and run it locally,” Babak noted. “And it’s not just, ‘Oh, it can do a bit of translation or summarization.’ The value is far greater.”

This is where things get exciting for entrepreneurs and specialists. DeepSeek didn’t just drop weights and walk away—they revealed their entire training process. That transparency lets other developers not only replicate their methods but apply them to existing models like LLaMA and push them even further.

Babak put it plainly:

“That’s a level of open sourcing that is very, very helpful. There’s a diversity of these models—from China, France, the US—that are open and can run locally. These are all good signs.”

And he’s right. This isn’t just about accessibility—it’s about shifting power. For the first time, smaller players can seriously compete. That’s not hype. That’s democratization.

So, What Exactly Is a Multi-Agent System?

Here’s where the conversation gets practical. Large language models (LLMs) are impressive, but they’re passive. They give you answers, but they don’t do. Agents, on the other hand, act.

“The difference in one word between an agent and a model is that an agent does stuff,” Babak explained. “Whereas the model just produces output.”

And agents aren’t limited to working solo. They can talk to each other, coordinate tasks, and, in time, form networks that span across entire companies. Imagine teams of digital specialists breaking down silos, sharing tasks, and scaling output without needing a bloated workforce.

“You can create extensive networks of these agents working with each other,” he said. “It breaks the silos of your organization, it can be very efficient, and it can improve quality on a lot of what you want to do.”

Here’s the link back to DeepSeek: lighter models mean running more agents becomes practical and affordable. And more agents means finer specialization, higher accuracy, and reduced risk if one model goes off track.

That’s not just technical progress—it’s a shift in how businesses can operate. It’s AI becoming less of a tool for giants and more of an everyday force multiplier.

Here’s a rewritten version that keeps the same structure but makes it more human, engaging, and tailored to tech experts, AI specialists, and business leaders:

A Platform Built for Choice

Enterprises don’t want to be boxed in by a single AI vendor—and honestly, who can blame them? Technology moves too fast, and nobody wants to rebuild their systems every time a new breakthrough drops. Cognizant saw that reality early on, which is why they created Neuro AI—a large-scale, open-source, multi-agent platform that’s deliberately model-agnostic.

Babak Hodjat put it simply:

“If tomorrow something better than DeepSeek or GPT-4.0 comes out, you shouldn’t have to start from scratch. You should be able to plug that new model into your agentic system and instantly level up its intelligence.”

This flexibility isn’t just about models. Neuro AI is also cloud-agnostic, giving businesses full control over where and how they deploy—whether for compliance, data residency, or privacy reasons. And the open-source approach? That wasn’t just a technical choice—it was a long-term strategy.

“We want Neuro AI to become the standard for enterprise-scale multi-agent systems,” Hodjat shared. “There’s now a thriving community outside of Cognizant actively building on and expanding the platform.”

Trust Isn’t Just UX—It’s the Foundation

Letting autonomous agents handle business-critical work is not just about speed or efficiency—it’s about trust. Trust that’s visible, measurable, and auditable. Cognizant has made that the centerpiece of Neuro AI.

Babak raised the hard questions many executives are already asking themselves:

“How do you measure trust? How do you earn it? And how do you design systems so you actually know how confident the AI is in what it’s doing?”

The answer lies in the separation between models and code. Yes, large models are inherently black-box. But the code that interprets their output and executes actions is fully transparent and under human control.

As Hodjat explained:

“The model may be opaque, but the code isn’t. We control the code. Which means we—the humans—always hold the authority.”

In practice, that means enterprise safeguards—like access controls, API permissions, and data tokens—sit firmly outside the model’s reach. The AI can reason, but it can’t bypass security rules. For enterprises, that isn’t just smart engineering—it’s non-negotiable.

Will Others Follow Cognizant’s Lead?

Babak hopes they will. He believes open source is the key to keeping AI innovation transparent, accessible, and sustainable—not just for enterprises, but for science itself.

“When breakthroughs are locked away behind closed doors and commercial walls, science suffers. Openness with a technology this powerful is the only path forward.”

Of course, not everyone shares that vision. Some companies that once championed openness in AI have pulled back—Meta being a notable example.

But there’s also momentum in the other direction. Babak points to the rise of open-source AI in China, where smaller, efficient models are being released freely.

“That’s good for humanity as a whole,” he said with conviction.

The Next Frontier: When Agents Start Talking to Each Other

So, where’s all this heading? Babak Hodjat is already spotting early signs of the next big shift. Picture this: not only do you have AI agents inside your company, but your customers – or even entire businesses – will have their own agents too. And those agents will start talking to each other. That’s the future unfolding right now.

“One thing that’s coming very soon is company agents communicating directly with agents that represent other companies, or even individual consumers,” Babak explained.

Of course, we’re not fully there yet. There are no global standards, no clear rules, and trust is still a huge question mark. But the momentum is undeniable. In fact, a major European bank recently asked Babak how they should even think about “marketing to an AI agent.”

At first, it sounds almost absurd. But step back for a moment: if an agent can evaluate pricing, check legal terms, compare offers, and even negotiate—well, at that point, it is the buyer. And if the buyer is an AI, companies will need sales strategies that speak the agent’s language, not just the human’s.

Modular AI: A Smarter Way to Build

Cognizant’s decision to design modular, LLM-agnostic platforms wasn’t about chasing trends—it was a response to the reality already in motion. Large language models are quickly becoming commodities. The real competitive edge comes from how you use them, connect them, and keep them under control.

“As long as your platform is modular, you can swap in a stronger model and expect it to perform the same job—just better,” Babak said.

That’s why architecture matters so much now. Open, flexible, and secure platforms aren’t just IT concerns anymore—they’re business decisions. AI isn’t a lab experiment; it’s becoming a teammate. And like any teammate, it needs context, guardrails, and collaboration to succeed.

The Bottom Line

What makes Babak’s perspective refreshing is his balance. He isn’t chasing science-fiction hype, nor is he leaning into doomsday fears. Instead, he’s calmly focused on how AI is practically evolving into something useful.

“We’re seeing these agentic systems begin to evolve on their own,” he noted. “They’ll know when to create a new agent for a task, or when two agents should merge to avoid duplication.”

This isn’t fantasy—it’s the roadmap. With open platforms, efficient models, and a strong focus on trust, that roadmap feels less like a dream and more like a reality finally taking shape.

FAQs

1. What’s the real difference between an AI model and an AI agent?

Think of an AI model as a specialist—it generates insights, predictions, or outputs based on data. An AI agent, on the other hand, is more like a problem-solver in action. It doesn’t just “answer,” it takes steps—coordinating with other agents, making decisions, and executing tasks within real systems. In short, models provide knowledge; agents put that knowledge to work.

2. Why does Cognizant lean toward open-source multi-agent systems?

Because trust matters. Open-source gives enterprises transparency into how things work, along with the flexibility to adapt and innovate without being tied down by vendor lock-ins. It’s about keeping control in your hands—while maintaining security and building confidence that your AI ecosystem is working for you, not boxing you in.

3. How could multi-agent systems transform the way businesses run?

Imagine your workflows running on autopilot, not in silos but across departments—and even beyond your company walls. Multi-agent systems make this possible. They can streamline operations, coordinate tasks, and even collaborate with agents from partner organizations. The result? Less friction, more agility, and a new kind of digital teamwork that changes how business actually feels day-to-day.