Transcript
organization and most of them will say no. Let's find out why. Hi, I'm Rohit Agarwal. Welcome to episode two of our series on building the AI control plane. In the previous episode, we discussed why enterprises need an AI control plane. We discussed how it helps discover AI usage, govern AI interactions, and secure AI agents from a unified platform. Today, we dig a little deeper into how the Prisma Airs AI Gateway helps discover AI usage in the enterprise. There are three key use cases we are seeing of AI in the enterprise. Coding assistants, enterprise agents, and co-pilots. 90% of the organizations use AI to assist with coding. Your engineering teams are using coding assistants every single day. You might know which coding assistants are allowed, but do you know what LLMs are these agents calling? Do you know what internal data they're pulling from your enterprise data systems? Do you know which tools and systems they're connecting to? And do you know how much of these interactions are actually costing you? The first problem is cost. A typical 50-turn AI assisted coding session can burn more than a million tokens. Multiply that across thousands of engineers and dozens of agents and you're looking at significant untracked spend. When you don't have visibility into usage, costs tend to spiral and you can't explain why. You can't point to which team, which agent, or which project is driving the increase in your AI bills. The second problem is traceability. In that same session, the agent might invoke dozens of tools, access multiple repos, and pull context from internal systems. But if you don't have visibility into these interactions, you have no audit trail. And the moment something goes wrong, a model hallucinates bad code into production, an agent accesses something it shouldn't have, you can't reproduce the data. You cannot answer the basic question, what happened? All of these problems come down to the same root cause. You do not have complete visibility. And without visibility, you're operating in the dark. To close this visibility gap, you place an AI gateway directly in the traffic path between your coding assistants and everything that's Now, every interaction flows through a single point. Every LLM call, every tool invocation, every connection to an internal system. All of it is logged, traced, and attributed. You can see usage broken down into three parts. The first part is the user interface. The second part is the API key. And the third part is the request and attribute it. You can see usage broken down by team, by user, by a single API key. And when the build spikes, you know exactly where to look. For auditability, now you have a complete record. Every request, every response, every tool your agent touched. If something goes wrong, you can create an action chain based on the traces logged and understand exactly what happened. What makes this even more powerful is this visibility is across all your AI. Your developers are using cursor, copilot, cloud code. Your product teams have custom built agents. Your support org is running enterprise copilots. Today, they all operate in silos. Each has its own usage patterns, its own costs, its own blind spots. The AI gateway brings all of that into one view. One place to see every agent, every model call, every tool interaction across every team in your organization. You go from scattered, fragmented, to a view with complete visibility. A single plane of glass for all AI activity. And that is the foundation the AI gateway is built on. Today, you learned how to answer an important question. Coding assistance activity across the enterprise. Next episode, we learn how to govern these interactions.
