Enterprise AI Moves Beyond Chatbots: Databricks Data Shows Rapid Shift to Agentic Systems

Enterprise adoption of artificial intelligence is undergoing a significant transformation, with businesses rapidly moving from experimental chatbot pilots to advanced, 'agentic' systems capable of orchestrating complex workflows across sectors. New insights from Databricks, which tracks activity across over 20,000 organisations—including 60 percent of the Fortune 500—highlight just how quickly this evolution is taking place.

The Agentic AI Revolution

The early hype around generative AI in the business world mostly yielded isolated chatbots and stalled proof-of-concept projects, often falling short of promised operational utility. However, the latest telemetry from Databricks suggests the sector has reached a turning point. A remarkable 327 percent growth in multi-agent workflows was observed on the Databricks platform between June and October 2025, indicating AI is fast becoming an embedded part of core system designs.

Supervisor Agents: Orchestrating Complexity

Fueling this shift is the rise of the 'Supervisor Agent'—a managerial AI component that orchestrates and delegates tasks among specialised sub-agents and tools. Since its debut in July 2025, this approach has quickly become the most common agent architecture, accounting for 37 percent of all multi-agent uses by October. Mirroring real-world managerial roles, supervisor agents handle intent detection, compliance checks, and routing of queries to domain-specific models or applications, significantly improving enterprise workflow automation.

While technology companies lead this adoption—building four times as many multi-agent systems as other industries—the applications extend far beyond Silicon Valley. For instance, financial services are employing multi-agent systems to simultaneously manage document retrieval and regulatory compliance, removing the need for human intervention in routine but critical business processes.

Traditional Infrastructure Feeling the Strain

As AI agents move from simple Q&A bots to autonomous actors executing business functions, the pressure mounts on underlying IT infrastructure. Systems originally designed for predictable, human-paced transactions now face high-frequency, programmatic operations. According to Databricks, while AI agents accounted for just 0.1 percent of database creation two years ago, they now make up 80 percent of such activity. Today, 97 percent of database test and development environments are spun up autonomously by AI agents—allowing teams to create temporary, purpose-built environments in seconds, enhancing agility across software development and data science.

The adoption of Databricks' platform for these tasks has surged, with over 50,000 AI and data applications created since the launch of Databricks Apps, and app creation growing by 250 percent in the past six months.

Embracing Multi-Model Strategies

Vendor lock-in remains a concern for many enterprise technology leaders, but the data reveals organisations are mitigating this through multi-model AI deployments. As of October 2025, 78 percent of companies use two or more Large Language Model (LLM) families—such as OpenAI's ChatGPT, Anthropic's Claude, Meta's Llama, and Google's Gemini. The number of enterprises leveraging three or more model families rose sharply from 36 percent to 59 percent between August and October 2025. This approach allows for smarter task allocation, with simpler jobs routed to smaller, more cost-effective models while saving the latest, most advanced models for complex reasoning problems.

Retail companies are notably at the forefront, with 83 percent using multiple LLM families to optimise both performance and cost. In this context, real-time AI is increasingly the norm: 96 percent of inference requests, according to the report, are performed in real-time rather than batch mode—a marked difference from the previous "big data" era. Industries such as technology and healthcare exemplify this with real-time-to-batch request ratios of 32:1 and 13:1, respectively. In fast-moving markets and applications like clinical support, even minor delays can result in significant lost value.

Governance: Acceleration, Not Obstruction

One of the more striking findings challenges the common notion that governance slows innovation. In practice, rigorous governance and evaluation frameworks are shown to accelerate production deployments in AI. Organisations utilising governance tools pushed 12 times more AI projects into production than those that did not. Similarly, systematic model evaluation led to nearly six times as many production deployments. Governance—covering areas such as data usage, compliance, and rate limitations—builds stakeholder confidence and helps move AI from experiment to scalable, safe deployment.

The 'Boring' Power of Automation

Despite the focus on AI's futuristic promise, its greatest current impact for businesses is automation of routine but essential activities. Databricks' data highlights several dominant use cases:

• Manufacturing and automotive: 35% focus on predictive maintenance.
• Health and life sciences: 23% center around summarising and synthesising medical literature.
• Retail and consumer goods: 14% involve market intelligence.

Moreover, 40 percent of top use cases address practical customer-facing concerns such as support and onboarding—a nod to immediate, quantitative business value.

As Dael Williamson, EMEA CTO at Databricks, notes, the conversation among businesses has "moved on from AI experimentation to operational reality." Williamson emphasises that those who are realising genuine competitive advantage are focusing less on the glitter of AI and more on engineering rigour, governance, and adoption of open, interoperable platforms. For organisations operating within Europe’s heavily regulated environments, such openness and control are not just desirable—they are decisive.

For the full article, see: Databricks: Enterprise AI adoption shifts to agentic systems.