Agentic AI Is Coming But Can Your Data Infrastructure Keep Up?
I teach a course on AI at the McCombs School of Business at UT Austin, where I regularly bring in guest speakers to share real-world perspectives with my students. Last September, I hosted DJ Patil — former U.S. Chief Data Scientist and now a prominent venture capitalist — and asked him what trend was drawing the most attention in Silicon Valley VC circles. Without hesitation, he said, “Agentic AI.” At the time, it was still under the radar for many, but just a few months later, interest in Agentic AI spiked sharply, confirmed by sensing mechanisms like Google Search Volume trends. What was once a forward-looking VC thesis is now rapidly becoming a defining shift in how AI will interact with and act on enterprise systems.
Ask five people what Agentic AI means, and you’ll likely get five different answers. Some define it narrowly, repackaging familiar use cases — like customer service chatbots or scheduling assistants — as “agents.” Others lean into more ambitious visions, where AI doesn’t just respond to prompts but acts autonomously, reasons across systems, and executes complex goals in dynamic environments. Our view aligns with the latter. At its core, Agentic AI refers to AI systems that can reason, plan, and act independently across multiple tools and data sources to achieve defined outcomes.
What Is Agentic AI and How Will It Change Business?
At its most practical, Agentic AI is about AI that does more than answer questions — it solves multistep problems.
Instead of responding to a prompt with a single answer, an agent might deconstruct a request into sub-tasks, query multiple data sources, take a series of actions across software tools, and report back on what it accomplished. Agents can analyze data, make API calls, update records, send notifications, ask follow-up questions, and even take consequential actions.
Agentic AI has the potential to drive meaningful transformation across critical business functions by acting autonomously across complex, disconnected systems. For example:
- Agentic AI will autonomously manage and optimize marketing campaigns by orchestrating tasks across disparate systems like ad platforms, web analytics, CRM tools, and content management systems. An agent might monitor real-time performance metrics, adjust bidding strategies, segment audiences, trigger creative updates, and even reallocate budget across channels based on what’s working, without human intervention. By continuously reasoning across these systems, it can test, learn, and adapt in real-time to maximize campaign impact.
- Agentic AI can transform logistics planning by coordinating across various systems, including inventory management, route optimization, carrier networks, and demand forecasting tools. An agent can monitor real-time supply and demand, adjust delivery routes, optimize warehouse allocation, and even shift resources or inventory between locations to avoid delays or stockouts. Reasoning across these interconnected systems enables more responsive, efficient, and autonomous logistics operations.
- Agentic AI can assist CFOs by integrating and reasoning across various financial systems, including ERP platforms, budgeting tools, revenue forecasts, and market data feeds. An agent can track cash flow in real-time, flag anomalies, model different budget scenarios, and even reallocate spending or adjust forecasts based on shifting business conditions. By continuously analyzing and acting on live financial data it enables faster, more informed decision-making at the executive level.
MCP: The Emerging Standard for Agent-Oriented APIs
To act autonomously, agents need to interface with a wide range of tools and data systems. That’s where Multi-Component Programming (MCP) comes in.
MCP is quickly emerging as the de facto standard for how agents interact with the software ecosystem. Inspired by traditional API-based integration, MCP allows AI agents to treat software systems not as static backends, but as dynamic, callable components in a broader plan of action.
In the same way that enterprises spent the last decade wrapping legacy systems with APIs to integrate them with modern applications, we are now seeing the same playbook applied to support AI agents. But this time, the interface isn’t just for developers — it’s for reasoning machines.
Enterprises are beginning to MCP-enable their systems so agents can:
- Fetch internal metrics
- Submit support tickets
- Update CRMs and ERPs
- Trigger automation pipelines
This shift is exciting, and it comes with profound implications for infrastructure.
Why Agentic AI Breaks Traditional Data Architectures
The power of Agentic AI lies in swarms of agents, not just one system, but dozens or hundreds operating in parallel to complete goals at scale.
And here’s the catch: most enterprise data systems were never built for this kind of concurrency. While your BI dashboard may support a few hundred daily users, agent swarms can generate thousands of queries per second (QPS) as they aggregate data, test hypotheses, and trigger decisions in real-time.
Consider an agent tasked with optimizing warehouse logistics. It might need to:
- Count inventory data
- Calculate historical shipping performance by region
- Check fleet availability
- Compare SLAs across providers
- Propose a route and update the fulfillment system
Now imagine 100s of those examples, with 1000s of agents doing this every few seconds. The resulting query volume would overwhelm a traditional data lakehouse.
This is the overlooked reality of Agentic AI: MCP exposes limitations in infrastructure — it doesn’t fix them. Agents are only as effective as the systems they rely on. If those systems can’t deliver real-time insights at high concurrency, agents stall, fail, or return stale data.
Agent-Ready Infrastructure
To support Agentic AI, enterprises need a data platform that’s:
- Built for low-latency, high-cardinality queries.
- Capable of handling massive concurrency without performance degradation.
- Flexible enough to serve both internal users and AI agents simultaneously.
This is where real-time OLAP systems like Apache Pinot come in. Built initially at LinkedIn to power real-time user analytics, Pinot gained traction as apps began embedding intelligence, driving a new kind of concurrency that traditional data systems couldn’t handle. As demand grew for live, interactive insights within customer-facing data products, Pinot emerged as the gold standard for powering these workloads.
Today, companies like Uber, Stripe, Walmart, and DoorDash rely on Pinot as a core part of their data platforms. These capabilities enable everything from fraud detection to dynamic pricing to personalized user experiences.
Now, as AI agents become the new apps, they bring with them even more aggressive demands on data infrastructure. Pinot’s architecture — optimized for subsecond queries, real-time upserts, and extreme QPS — makes it uniquely suited to meet these demands. It handles billions of rows, supports complex filtering and aggregations, and delivers insights with the speed, scale, and freshness that modern agents need to reason, plan, and act in real-time.
Agentic AI is not a trend but a transformation. As AI agents become central to how work gets done, how customers are served, and how businesses operate, the supporting infrastructure must evolve.
