Turning AI Chaos into Control: A Conversation on Agentic AI with Tesseract Talks

As enterprises move from experimenting with large language models to deploying agentic AI systems in production, a new set of challenges is emerging. Teams are moving faster than ever, but often in different directions. Models, tools, frameworks, and agents are multiplying, and with that growth comes fragmentation.

In a recent episode of Tesseract Talks, John K. Thompson sat down with Nikunj Bajaj, Co-founder and CEO of TrueFoundry, to explore what it really takes to scale agentic AI inside large organizations.

Here are some of the most important takeaways from the conversation.

From Simple LLM Apps to Complex Agentic Systems

AI systems have evolved drastically, especially over the past year.

What used to be a single LLM call with a tool has now become a network of components working together. Production-grade agents typically combine:

• Multiple LLMs (often across vendors and clouds)
• Model Context Protocols (MCPs) and tools
• Guardrails for safety, privacy, and compliance
• Prompts and orchestration logic
• Other agents, composed hierarchically

As Nikunj explained, this complexity isn’t accidental, it’s a natural result of agents becoming more capable. But it also means that different teams inside the same enterprise are building agents in very different ways, using different stacks and frameworks.

That flexibility helps teams move fast. At scale, it also creates chaos.

The Real Enterprise Challenge: Speed and Control

Enterprises are faced with the need to balance two competing forces: giving teams autonomy to experiment and deliver value quickly while at the same time, maintaining enterprise-wide consistency around security, governance, and cost Nikunj framed this as “federated execution with centralized governance”.

Early on, when teams are small, autonomy works well. But as organizations grow, policies, budgets, and oversight become essential. Human employees operate with flexibility but within a structured system. Agentic AI needs the same thing.

Agentic AI needs the same thing.

This is where the idea of an AI Gateway comes in. According to Nikunj, the gateway has evolved far beyond a simple proxy for routing requests between models. Today, it is becoming:

• A unified entry point for LLMs, MCPs, agents, prompts, and guardrails
• A normalization layer across cloud providers and model vendors
• A place to enforce access control, budget limits, and compliance policies
• A foundation for observability and debugging

In short, it becomes the agentic headquarters, the control plane that enterprises have been missing. 

How We Think About TrueFoundry’s Role

At TrueFoundry, we don’t see agentic AI as a series of one-off projects. We see it as a long-term transformation. As Nikunj explained, our focus is on helping enterprises:

• Keep their AI stack future-ready as the ecosystem evolves
• Integrate new agentic capabilities into existing infrastructure
• Avoid the false choice between building everything themselves and buying rigid tools

By providing a flexible, API-driven platform, teams can build on top of a strong foundation and move faster without losing control.

As Nikunj put it, “Agents need flexibility to act. Enterprises need a headquarters to control them.”