Guest: Varun Talwar
Company: Tetrate
Show Name: An Eye on AI
Topic: AI Governance

The enterprise AI conversation is stuck on models. But the real friction point for production deployments isn’t about which LLM to use—it’s about trust, governance, and the connectivity layer that makes agents safe to wire into business systems. Two recent moves from Tetrate signal where the industry needs to go: the launch of the Agentic AI Foundation with Tetrate as a founding Gold sponsor, and the debut of Agent Router Enterprise. Both point to a future where secure, governed connectivity becomes the backbone of AI systems.

Varun Talwar has seen this movie before. As CTO of Tetrate and an early contributor to CNCF projects like Envoy and gRPC, he watched Kubernetes go from experimental to ubiquitous. Now he sees agent AI moving three times faster—but with a critical infrastructure gap.

The fundamental shift happening now is workload behavior. Human-generated traffic is predictable, low-concurrency, request-response. Agent-generated traffic is recursive, bursty, and massive. A single agent request to refactor a codebase or fix security logs can fan out to thousands of backend systems. Talwar describes it bluntly: “It doesn’t feel like a normal request. It feels like a DDoS attack.”

Legacy enterprise APIs were architected for user interaction, not agent recursion. They lack the control plane, security posture, and identity enforcement needed for agent-native architectures. This mismatch is why many organizations are stuck in experimentation mode.

Why the Agentic AI Foundation Matters

The Agentic AI Foundation launched this week with Tetrate as a founding Gold sponsor. For Talwar, the timing reflects industry maturity. Open source accelerates innovation by bringing the energy of multiple companies and communities together. The Foundation’s first major project is the Model Context Protocol, donated by Anthropic and now under neutral governance.

MCP is a standardized way for models to query backend tools, APIs, and data sources to get the context they need. It eliminates fragmentation and pre-embeds governance and security into the spec. Tetrate has been working in this space for over a year, making Envoy and its AI Gateway MCP-ready, and building services like Agent Router Enterprise to enhance MCP security further.

The Security Layer No One Talks About

Security is the number one blocker for enterprise AI deployment. Talwar shared an example: connecting Claude to Google Drive requires permissions to read, write, and delete files. In a G Suite enterprise environment, that inadvertently grants an AI prompt access to all corporate data. One poorly written prompt could leak sensitive information.

This isn’t theoretical. C-level executives across industries are asking the same question: How do I enable innovation without creating security holes? Agent Router Enterprise addresses this by overlaying stricter access controls, confidentiality checks, and identity enforcement on top of MCP. The goal is to make it impossible to accidentally leak data or grant unauthorized access, even with imperfect prompts.

Beyond Security: Quality and ROI

Once organizations solve the security problem, the next question is quality. Talwar recently attended a summit at one of the largest banks where they had deployed AI for code generation. The statistics showed faster code production, but leadership was asking: Is it maintainable? Is it business-impacting? Can we support and extend it?

Then comes ROI. AI infrastructure isn’t cheap, and enterprises need proof of value. These are harder problems that require the entire industry to collaborate. Talwar believes the Agentic AI Foundation provides the umbrella for those conversations to happen faster.

Making MCP Production-Ready

Tetrate’s work on MCP goes beyond implementation. Last week, Talwar’s team met with Anthropic and MCP maintainers to discuss tracing metadata gaps in the spec. The response was immediate: form a working group, submit an extension proposal, and get it into the next MCP release. That’s the kind of velocity open governance enables.

Talwar’s ideal vision for the Foundation is simple: multiple working groups making incremental progress across SDKs, governance, registries, and traceability. With enough end-user collaboration, the industry can ensure it’s solving the right problems at the right pace.

What This Means for Enterprises

If your organization is experimenting with AI agents, the shift to production requires rethinking backend infrastructure. APIs need to become modular, MCP-ready tools with built-in identity and security. The connectivity layer can no longer be an afterthought—it’s the foundation for trust at scale.

Tetrate’s dual announcement—joining the Agentic AI Foundation and launching Agent Router Enterprise—represents a bet that the next wave of enterprise AI will be defined not by models, but by the infrastructure that makes those models safe, governed, and interoperable.