AI Investment Meets Enterprise Reality
Enterprise AI adoption is entering a more practical phase. Companies are looking beyond hype, market speculation and infrastructure buildouts. Roman Stanek, Founder and CEO of GoodData.AI, joins Techstrong.ai to explain why an AI bubble is not always bad for enterprise leaders. Bubbles often overbuild useful infrastructure. The dot-com era left behind fiber capacity that later powered new waves of innovation.

The challenge is that enterprise AI adoption will not move at consumer-app speed. Enterprises have complex operations, slow decisions and years of data issues. AI demands a different operating model. It rewards agility, experimentation and fast iteration. Many enterprise IT organizations are still built around long planning cycles and predictable deployments.
AI-Native Workflows Need a New Mindset
Stanek warns that inserting agents into existing workflows may only add more complexity. Enterprises need to rethink how work gets done in an AI-native environment. That means questioning innovation committees, legacy transformation models and old staffing assumptions. It also means giving smaller teams room to move faster.

The conversation highlights a deeper tension. AI agents can be probabilistic, fast-moving and hard to govern with old models. Yet Stanek says the bigger issue is organizational agility. Models, tools and development patterns can change daily. Quarterly innovation meetings cannot keep pace with that rate of change.
Data Readiness Still Comes First
GoodData.AI focuses on helping organizations organize data for AI. Stanek says many companies still need to revisit core data management practices. Enterprise data may be abundant, but it is often not ready for agents. It also may not support semantic models or governed AI applications. Without reliable structures, AI projects can struggle to produce useful outcomes.

That is why the discussion turns from abstract agentic AI to practical foundations. Teams need transformations, semantic layers, metadata and shared business meaning. Stanek sees an abundance of intelligence emerging. Enterprises still need better ways to deploy it securely, predictably and economically.
Small Models and Applications Shape What Comes Next
The episode also explores whether smaller models and local inference could shift AI economics. Stanek says enterprise tasks often do not require the largest frontier models. Many use cases involve focused business optimization. Smaller models can deliver stronger price performance for those workloads.

For IT leaders, the most important decisions may not be about GPU scarcity. Stanek points instead to the application layer. Enterprises need to decide what they will build and who will build it. They also need to decide how AI can change the business. The next phase of enterprise AI will be defined by practical applications, trusted data and faster innovation without losing control.