AI agents are reshaping how enterprises interact with data, but they bring a stark warning: closed data ecosystems are crumbling under the load. In a recent episode of The New Stack podcast, Fivetran's Chief Product Officer, Anjan Kundavaram, laid out the economic and operational challenges that arise when agent-driven queries—often ten to a hundred times more numerous than human-era analytics—collide with inflexible, siloed stacks. Drawing from a conversation at Google Cloud Next, Kundavaram argues that the only path forward is open data infrastructure. Here are the ten essential takeaways from his analysis, each explaining why locked-down systems are doomed in the age of autonomous agents.
1. Agents Multiply Query Volume Like Never Before
An AI agent operating on a data warehouse can fire off ten to a hundred times more queries than a human analyst ever could. This explosion in query volume is not a bug—it's the feature of automation. But it puts immense pressure on data architectures. In closed ecosystems, every single query is forced through the same premium compute path, turning what should be a cost efficiency into a budget disaster. Kundavaram likens this to "using a Lamborghini to mow the lawn all the time"—overkill that drains resources fast. The volume itself isn't the problem; the inescapable routing to expensive engines is.
2. Agents Don't Need Instant Answers—and That's a Superpower
Unlike humans, agents are patient. They can wait seconds or even minutes for an answer if it saves significant cost. As Kundavaram notes, "An agent could go spend more time if the agent thinks you're going to save 10x the cost." In an infrastructure with heterogeneous compute tiers—some fast and pricey, others slower but cheap—agents can intelligently route the query to the right engine. A complex analytical question goes to the heavy lifter; a simple check goes to the lightweight option. Closed stacks obliterate this flexibility, forcing every query through the same expensive door.
3. Multiple Compute Engines Unlock Cost-Effective Query Routing
The key to optimizing AI-driven workloads is having multiple compute engines at your disposal. An open data stack enables agents to choose the best engine for each query based on cost, latency, and complexity. This routing capability is a game-changer: expensive analytical queries get the horsepower they need, while cheap, frequent queries run on low-cost infrastructure. Closed stacks, by contrast, present a single, toll-booth gateway. Kundavaram warns that this lack of choice will rapidly erode analytics budgets as agent usage expands.
4. Data Silos Create a 'Triple Whammy' of Bad Outcomes
When enterprise data lives in dozens of disconnected systems, the AI doesn't have a unified context. Kundavaram describes the consequences as a "triple whammy": poor AI answers (because the model lacks complete information), sharply higher costs (because agents run far more queries trying to piece things together), and massive waste (because those queries are fed with weak context). Consolidating data into a single, open lake or warehouse is not just a convenience—it's a necessity to avoid this downward spiral.
5. The Instinct to Clamp Down on Query Budgets Is Wrong
Faced with rising query costs, many data leaders immediately tighten controls. Kundavaram recounts a conversation with an analytics leader at a large company who said, "Our analytics budgets, just queries, have gone up a lot. I got to put control." His response? "No, no, don't put controls. Let's innovate." The reflex to lock down access kills the very productivity gains that agents promise. Instead of restricting queries, organizations should invest in smarter infrastructure that allows agents to self-optimize.
6. Open Infrastructure Enables the Agent's Cost-Saving Instinct
Agents are naturally designed to minimize cost if given the option. But they can only do that if the underlying data stack is open—meaning it exposes multiple query engines, transparent pricing, and semantic metadata. Fivetran's Open Data Infrastructure framework aims to create exactly this environment. Kundavaram emphasizes that the "productivity unlock from agentic analytics only materializes if customers refuse the lockdown instinct and invest in open infrastructure and semantic discipline."
7. Fivetran's Data Access Benchmark Exposes Hidden Vendor Taxes
At Google Cloud Next, Fivetran launched an Open Data Infrastructure Data Access Benchmark. Its purpose: to make it harder for vendors to quietly mark up costs on AI workloads by locking data behind proprietary compute paths. The benchmark provides a transparent way to compare the true cost of running agent queries across different stacks. It's a move designed to give customers leverage and encourage ecosystem openness. As Kundavaram sees it, "The argument has timing on its side."
8. Most Enterprise Data Systems Were Never Built for Agent Swarms
Earlier this year, The New Stack reported that the majority of enterprise data architectures were designed for human-driven, batch-oriented workflows—not for hordes of autonomous agents making simultaneous queries. This mismatch explains why so many organizations are suddenly seeing budget overruns and performance issues. Agents expect a modern, scalable, cost-aware environment; closed stacks deliver the opposite. The gap is forcing a rethinking of data strategy from the ground up.
9. Semantic Discipline Is as Important as Infrastructure
Open infrastructure alone isn't enough. Consistent, well-defined semantics—common definitions for terms, data types, and business rules—are critical for agent reasoning. Without them, agents can misinterpret data or generate inconsistent results, even when the compute engines are open. Kundavaram urges companies to invest in semantic layers that provide a single source of truth for AI consumption. This discipline amplifies the benefits of open stacks and prevents the "triple whammy" from recurring at the metadata level.
10. The Future Belongs to Open Data Ecosystems
Kundavaram's overarching message is clear: closed data stacks won't survive the agent era. The economic pressure—more queries, higher costs, scattered context—will force organizations to open up or face severe competitive disadvantages. Fivetran, with its data lake interoperability work on Google Cloud and its benchmark initiative, is betting that openness will become the default. For enterprise leaders, the choice is stark: evolve toward an open, agent-friendly architecture or watch your analytics budgets spiral out of control.
The agent era is not a distant scenario—it's already here. As Kundavaram's insights reveal, the companies that thrive will be those that embrace open data infrastructure, semantic consistency, and a mindset of innovation over restriction. The Lamborghini belongs on the racetrack, not the lawn.