Nikunj Bajaj, Co-Founder and CEO, TrueFoundry chats about the evolving conversations (and concerns!) around AI with insights on what makes for optimized deployment of AI tools and agents in this AiThority interview:
______
Hi Nikunj, tell us more about your latest acquisition of Seldon AI and how it changes things for the end users of your platform?

Seldon spent a decade solving one of the hardest problems in production AI: serving ML models in real time, at scale, and on Kubernetes for enterprises like Best Buy, Verizon, and PayPal. Those same enterprises are now being asked to ship agentic AI, and they don’t want a second stack, a second vendor, or a second governance model to do it.

This acquisition gives them one platform for both. Seldon’s customers keep their production ML running on our AI Deploy platform, and they get a clear path into agents, LLMs, and tools through our AI Gateway, all on the Kubernetes foundation they already trust. For TrueFoundry’s customers, nothing changes, but the deep vertical expertise Seldon’s customers bring in banking, retail, telecom, insurance and healthcare makes the platform better for everyone. A decade of ML investment becomes a launchpad for agentic AI, not something to walk away from. That’s the whole point.

Also Read: AiThority Interview with Gou Rao, co-founder and CEO at NeuBird AI

Agentic AI and enterprise AI: what’s trending today?

The conversation has shifted from which model to use toward how to actually operate systems once dozens of agents are running in production. A year or two ago, most organizations were dealing with one model and one use case, so the questions were relatively simple.

Not anymore. Agents now touch real systems and real data across multiple teams, and the hard question isn’t about model selection. It’s operational. How do you keep visibility and control as dozens of agents run in production without grinding everything to a halt?

What’s emerging is consolidation. Fewer disconnected tools, more unified oversight of what every agent is doing, what it’s accessing, and what it costs.

What are some of the biggest misconceptions around agentic AI that you’d like to address in this conversation?

The biggest misconception is that choosing the right model is the hard part of agentic AI, but the hard part starts the moment an agent gets access to real systems, real data, and the ability to take action. That introduces a different failure mode than anything a standalone model produces.

We recently surveyed over 200 enterprise AI leaders running agents in production and found that 76% lack unified logging across their models and agent workflows, and 56% have no centralized control or governance layer at all.

There’s another common assumption that governance and autonomy are in tension and that adding guardrails necessarily slows agents down. In practice, it’s closer to the opposite. Strong observability and control are what make it possible to extend more autonomy to an agent because you can verify what it’s doing rather than hoping it behaves as expected.

What best practices should business heads and operational teams follow when deploying agentic AI across different functions?

The first step is establishing visibility. Teams need to know which agents exist and what systems and data they can access before expanding their scope of responsibility.

Second, organizations should build on the infrastructure they already operate rather than standing up a parallel stack for agentic workflows, since that’s often where operational cost and fragmentation come from.

Third, bring in the people accountable for outcomes, not just the AI team, from day one. Agentic AI touches operations, compliance, and the business simultaneously, so all three need a seat at the table from the start.

As agentic AI becomes more mainstream, where do you see the balance between business operations and AI shifting?

That distinction is going away. ‘The AI team’s project’ and ‘how the business actually runs’ are becoming the same sentence. Agentic AI is moving from a bounded experiment into core operational workflows, which means the AI infrastructure team needs to understand the operational consequences of what they build, while operational leaders need enough visibility into these systems to actually trust and audit them. Both sides have to meet in the middle.

What are a few cautionary aspects regarding AI that you’d share with our readers before we wrap up?

Organizations should be careful not to let the pace of deployment outpace visibility and control.

Every agent put into production without proper oversight introduces a liability that isn’t visible until something goes wrong. This is similar to how partial failures or silent degradations in a system are often harder to catch than a hard outage.

There’s also a cost to fragmentation. Running multiple models and agent frameworks without a unified view of activity and spend tends to produce unpredictable costs and no clear audit trail.

Also Read: ​​AI and The Future of Work: Artificial Intelligence Is Expanding Organizational Intelligence Beyond Human Limits

[To share your insights with us, please write to psen@itechseries.com]

TrueFoundry is an Enterprise Platform as a Service that enables companies to build, observe, and govern Agentic AI applications securely, scalably, and with reliability through its AI Gateway and Agentic Deployment platform. Leading Fortune 1000 companies trust TrueFoundry to accelerate innovation and deliver AI at scale, with over 1 trillion tokens per day processed via the TrueFoundry AI Gateway and more than 1,000 clusters managed by its Agentic

Nikunj Bajaj is Co-Founder and CEO of TrueFoundry, an AI infrastructure platform helping enterprises deploy, govern, and scale AI in production. Before founding TrueFoundry, Nikunj led conversational AI at Meta, where he built the ML systems powering Facebook Messenger for over a billion users and shipped Meta’s first on-device deep learning model. Under his leadership, TrueFoundry has raised $21M from investors including Intel Capital, PeakXV, Eniac, and Jump Capital, and is trusted by organizations like Pfizer, Siemens Healthineers, and Fortune 1000 enterprises.

deployment platform. TrueFoundry’s vision is to become the central control plane for running Agentic AI at scale within enterprises, serving as the command center for enterprise AI. Headquartered in San Francisco, TrueFoundry operates across North America, Europe, and Asia-Pacific, supporting enterprise AI deployments for some of the world’s most innovative organizations.