7 Reasons Closed Data Stacks Will Fail in the Age of AI Agents

The arrival of AI agents is reshaping how organizations interact with their data. These autonomous programs don’t just run a few queries—they can execute ten or a hundred times more than any human-driven analytics workflow ever did. For closed data ecosystems, this shift spells trouble. As Anjan Kundavaram, Chief Product Officer at Fivetran, explains in a recent interview, closed stacks route all that query traffic through the same expensive compute path, leading to skyrocketing costs and subpar results. In this article, we break down seven critical insights from Kundavaram’s arguments, drawn from his conversation at Google Cloud Next. Whether you’re a data leader or an AI strategist, understanding these points is vital for building infrastructure that can thrive in the agent era.

1. Agents Run 10 to 100 Times More Queries Than Humans

The most fundamental change is the sheer volume of queries. An AI agent, left to explore a data warehouse, can generate hundreds of times more requests than a human analyst ever would. This isn’t a bug—it’s a feature of autonomous decision-making. But it puts immense pressure on data systems that were designed for occasional human queries. As Kundavaram notes, “An agent could go spend more time if the agent thinks you’re going to save 10x the cost.” The implication is clear: the old assumption of low query volumes no longer applies. Organizations must plan for this surge or risk system overload and budget blowouts.

2. Closed Stacks Force Every Query Through Expensive Compute

In a closed data stack, all queries—whether simple or complex—travel through the same costly compute engine. Kundavaram likens this to “using a Lamborghini to mow the lawn all the time.” The problem is that agents don’t distinguish; they send every request down the same pipeline, wasting resources on trivial operations that could be handled cheaply. This inefficient routing is a direct consequence of vendor lock-in, where the stack lacks flexibility to route queries to more appropriate, lower-cost engines. Open infrastructure, by contrast, allows dynamic resource allocation, saving money and improving performance.

3. Agents Can Optimize Costs—But Only With Multiple Engines

Kundavaram highlights that agents are capable of intelligent cost optimization—if the underlying infrastructure supports it. With a stack that offers multiple compute engines, an agent can send complex analytical questions to a powerful engine and simple ones to a lightweight, cheaper option. This is the core benefit of an open data architecture. In a closed stack, however, every question goes through the same expensive door, eliminating any chance of cost savings. As agents grow more sophisticated, the ability to route intelligently becomes essential for sustainable AI operations.

4. The Triple Whammy: Bad Answers, High Costs, Wasted Context

When data and context are scattered across many systems, the consequences compound. Kundavaram describes a “triple whammy”: first, AI answers are poor because the agent lacks consolidated context; second, costs spike because agents run far more queries to compensate; and third, those queries are fed with weak context, wasting compute cycles. This vicious cycle can quickly erode the value of agentic analytics. The solution, he argues, is to consolidate data into a unified, open platform before deploying agents. Otherwise, you’re paying more for worse results.

5. The Wrong Reflex: Clamping Down on Queries

Faced with surging query costs, many data leaders instinctively tighten controls. Kundavaram recounts a conversation with a data leader from a large company who saw analytics budgets skyrocketing and immediately wanted to limit queries. But Kundavaram’s advice was clear: “Don’t put controls. Let’s innovate.” The reflex to clamp down stifles the very productivity gains that agents promise. Instead, leaders should embrace the explosion in query volume as a signal to invest in more efficient infrastructure, not to restrict access. Innovation requires experimentation, and that means letting agents run freely within an optimized environment.

6. The Right Move: Invest in Open Infrastructure

To unlock the potential of agentic analytics, Kundavaram prescribes a shift to “Open Data Infrastructure.” This means moving away from closed, proprietary stacks that impose vendor lock-in and toward systems that support multiple engines, open formats, and interoperability. Fivetran has been championing this approach with its Open Data Infrastructure framework and a new Data Access Benchmark that exposes hidden costs in vendor pricing. The goal is to make it harder for vendors to quietly tax AI workloads. For organizations, this investment pays off through lower costs, better answers, and future-proof scalability.

7. Semantic Discipline Is the Key to Productivity Gains

All the open infrastructure in the world won’t help if your data lacks semantic consistency. Kundavaram emphasizes that the productivity unlock from agentic analytics only materializes when customers enforce semantic discipline—standardizing definitions, metadata, and context across the data landscape. Without this, agents will continue to produce unreliable results, even on a well-architected stack. Thus, the final piece of the puzzle is investing in data governance and semantic layers. Combined with open infrastructure, this allows agents to deliver on their promise of transformative efficiency.

In conclusion, the era of AI agents demands a fundamental rethinking of data architecture. Closed stacks that served human analysts well are now liabilities, forcing costly inefficiencies and poor outcomes. By embracing open infrastructure, allowing intelligent query routing, and maintaining semantic discipline, organizations can turn the agent wave into a competitive advantage. As Kundavaram puts it, the key is to innovate rather than clamp down. The future belongs to those who build for openness, scalability, and intelligent cost management. For more insights, revisit Reason 1 or Reason 6 to see how each element fits into the bigger picture.