A new industry report has put a number on something most data and infrastructure leaders have felt in their gut for a while: the infrastructure most enterprises run today was not built for what agentic AI actually demands of it. According to the report, 83% of surveyed organizations believe their current infrastructure requires some level of upgrade to support production-grade agentic AI systems — and only 17% report full confidence in supporting mission-critical, production-grade agents.
Break down that 83% and the picture gets more specific: 12% say they need significant fundamental upgrades, 29% report needing major upgrades to specific core systems, and 27% describe the gap as minor integration work and tuning. Only 16% say their infrastructure can support initial pilot agents with minimal effort. In other words, the gap between “we ran a promising agent demo” and “we can run this agent in production, reliably, at scale” is enormous — and it’s an infrastructure problem as much as it’s a model problem.
Why agentic AI breaks assumptions your data platform was built on
The report’s underlying diagnosis is worth sitting with, because it explains why this gap exists in the first place. Traditional analytics workloads are relatively well-behaved: a dashboard refreshes on a schedule, a report runs at a predictable time, a data pipeline executes in a known window. Agentic AI workloads don’t play by those rules.
A single agentic prompt can initiate hundreds of downstream actions as an agent independently browses, queries, and executes across multiple systems. And unlike a traditional query, that workload is persistent, stateful, and potentially long-running — the agent needs to maintain context across an entire workflow, not just return a result set and disappear. That means the infrastructure underneath it has to do several things at once: provide access to many different data sources, sustain continuous execution without degrading, and maintain security and governance controls the entire time — not just at the query boundary.
Critically, the report also calls out that this has to work across on-premises, edge, and cloud environments simultaneously — because enterprise data, and the systems agents need to interact with (ERPs, CRMs, data warehouses, operational databases), rarely live in one place. An agent that can only “see” the data that happens to sit in one cloud region isn’t a production-grade agent. It’s a demo with a narrower blast radius.
This is an architecture problem, not a model problem
It’s tempting to read a stat like “83% need an infrastructure upgrade” and assume the fix is a bigger model, a faster GPU cluster, or a more sophisticated orchestration layer. Those things matter, but they’re not the bottleneck this report is describing. The bottleneck is the data layer underneath all of it — the thing that has to serve up governed, secure, low-latency access to enterprise data across every environment an agent might need to touch, at a concurrency level no BI dashboard ever had to sustain.
This is exactly the architectural gap a hybrid, high-performance SQL data platform is designed to close. Instead of forcing every agent interaction through a single cloud environment, or bolting governance onto the edges of a system that wasn’t built for continuous, high-concurrency access, the right foundation gives agents consistent, secure, high-performance access to data wherever that data actually lives — on-premises, in the cloud, or both — without asking the business to choose one environment and live with its limitations.
Enterprises racing to move agentic AI from pilot to production would do well to ask a blunt question before they scale up the next proof of concept: is the gap really in the agent, or is it in the data foundation underneath it? For the 83% in this report, the honest answer is probably the latter — and no amount of prompt engineering fixes an infrastructure problem.
