An operating model for enterprise AI — not model selection or tooling — is what separates organizations making sustained progress from those stuck in experiments. That is the central argument of an interview published on the Databricks blog with Dael Williamson, the company’s chief technology officer for EMEA. As AI becomes embedded in core business processes, the executive question is shifting from whether AI delivers value to whether the organization is structured to support it over time.
Signs an organization is serious about AI
Williamson’s first indicator is ownership: who owns the data, who owns the AI, and how close that ownership sits to the CEO. Ownership near the top signals strategic importance. In practice, data and AI are often owned by different groups at different levels, and when AI is structurally disconnected from data, the result is stagnant use cases and fragmented experiences — a problem in a world where traffic, markets, and supply chains shift constantly.
The second indicator is whether the organization maintains an inventory of its data assets. Financial and physical assets are meticulously documented, yet many organizations do not fully know what data they hold, where it resides, or what it is worth. The third is how broadly data is defined: beyond structured tables and logs, images, emails, collaboration documents, and code all carry operational insight, and organizations that widen their definition unlock far more value.
Why proximity between data and AI matters
When data and AI share a foundation, more dynamic use cases become possible; when they are separated, AI depends on slower, more static inputs. Traditional governance and catalog tools handle structured data well but struggle with unstructured, fast-changing sources — one reason comprehensive data inventories remain rare. For problems such as liquidity modeling, credit risk, or supply chain resilience, AI must work directly with timely, continuously updated data, or insights arrive after the moment they were most useful.
Connecting central teams and the business
Williamson argues the leader responsible for data and AI needs a seat at the executive table and a genuine understanding of how these systems behave, because AI does not behave like traditional software. On tooling, leading companies resist relying exclusively on the AI features embedded in dozens of SaaS products: those features boost individual productivity but rarely help teams work cohesively, and over time they entrench inconsistent definitions, metrics, and processes.
The build-versus-buy calculus is also being rethought. The goal is not building everything in-house, but avoiding excessive lock-in — portability, transparency, and control over data and AI assets are gaining weight. And winning organizations manage AI initiatives as a portfolio of bets rather than a linear roadmap: some projects get stopped, others earn further investment, and the portfolio adapts as technology and conditions evolve.
The three-year outlook
Looking ahead, Williamson expects the traditional separation between IT and business to keep eroding: business teams become more technically adept, technical teams align more closely with outcomes. IT’s historical strongholds — risk management, governance, operational complexity — are areas where AI is increasingly effective, particularly cybersecurity, IT support, and compliance. As legacy complexity shrinks and siloed vendor ecosystems recede, teams will be defined less by the systems they run and more by the results they deliver, potentially giving rise to entirely new lean units focused on value creation.
How skills and roles evolve
Many IT organizations will continue to shrink as decades-old systems are retired, while the software development lifecycle itself changes: manual coding is increasingly AI-assisted, and effort shifts toward evaluation, behavior testing, guardrails, and ongoing monitoring. New roles are emerging around observability, orchestration, and system inspection, blending technical, operational, and organizational skills — and not always coming from traditional engineering backgrounds. Management evolves too, refocusing on analysis, decisions, and workflow improvement as AI absorbs administrative tasks. Critical thinking, experimentation, and an analytical mindset become the durable skills.
Limitations and what to watch
Context matters when weighing these views: they come from a senior executive at a data-and-AI platform vendor whose commercial interest lies in unified data platforms, so the emphasis on consolidating data and AI on one foundation should be read alongside independent perspectives. Timelines for IT reorganization are also notoriously hard to predict, and heavily regulated industries may move far more slowly than the three-year sketch suggests. Still, the core diagnostics — executive ownership, data inventories, portfolio management of AI initiatives — echo advice from many independent consultancies. A concrete example of the collaborative, evaluation-first workflow described here appears in this walkthrough of building a regulatory extraction agent.
