Enterprise AI is undergoing a profound transformation, shifting its focus from basic generative applications to advanced agentic systems, according to recent data from Databricks. While the initial wave of generative AI offered promises of business overhaul, its practical implementation often stalled at rudimentary chatbots or unscaled pilot programs. However, fresh telemetry compiled by Databricks suggests a turning point has arrived in the market.
Databricks’ analysis, encompassing over 20,000 organizations including 60 percent of the Fortune 500, reveals a rapid movement towards “agentic” architectures. In these new setups, AI models do more than just retrieve information; they independently plan and execute intricate workflows.
The Ascent of Multi-Agent Workflows
This evolution necessitates a fundamental reallocation of engineering efforts. Between June and October 2025, the utilization of multi-agent workflows on the Databricks platform surged by an impressive 327 percent. This significant increase signals that AI is transitioning from a supplementary tool to a foundational component of system architecture.
A key driver behind this expansion is the ‘Supervisor Agent’. Instead of relying on a singular model to address every request, a supervisor functions as an orchestrator. It disaggregates complex queries and assigns tasks to specialized sub-agents or specific tools. Since its introduction in July 2025, the Supervisor Agent has become the foremost agent use case, accounting for 37 percent of total agent usage by October. This operational pattern mirrors human organizational structures, where a manager oversees tasks rather than performing each one personally, ensuring team execution. Similarly, a supervisor agent manages intent detection and compliance before directing work to relevant domain-specific tools.
Technology firms are currently leading this adoption trend, developing nearly four times more multi-agent systems than any other sector. Nevertheless, the benefits extend across diverse industries. For example, a financial services entity could leverage a multi-agent system to concurrently manage document retrieval and ensure regulatory compliance, thereby delivering verified client responses without human intervention.
Evolving Infrastructure Requirements
As AI agents advance from answering questions to actively executing tasks, the underlying data infrastructure faces new demands. Traditional Online Transaction Processing (OLTP) databases were designed for human-speed interactions, characterized by predictable transactions and infrequent schema modifications. Agentic workflows fundamentally challenge these assumptions.
AI agents now generate continuous, high-frequency read and write operations, frequently establishing and dismantling environments programmatically for code testing or scenario simulations. The sheer scale of this automation is evident in telemetry data: two years ago, AI agents created merely 0.1 percent of databases; today, that figure stands at 80 percent. Furthermore, 97 percent of database testing and development environments are now provisioned by AI agents. This capability allows developers to launch ephemeral environments within seconds, rather than hours. Over 50,000 data and AI applications have been developed since the Public Preview of Databricks Apps, experiencing a 250 percent growth rate over the past six months.
The Multi-Model Approach and Real-Time Imperative
To mitigate the persistent risk of vendor lock-in, enterprises are increasingly adopting multi-model strategies for agentic AI. Data from October 2025 indicates that 78 percent of companies utilized two or more Large Language Model (LLM) families, such as ChatGPT, Claude, Llama, and Gemini. The sophistication of this strategy is growing, with the proportion of companies employing three or more model families rising from 36 percent to 59 percent between August and October 2025. This diversity enables engineering teams to route simpler tasks to more cost-effective models while reserving advanced models for complex reasoning. Retail companies are particularly adept, with 83 percent employing multiple model families to balance performance and expenditure. A unified platform capable of integrating various proprietary and open-source models is fast becoming an essential component of the modern enterprise AI stack.
Contrary to the legacy of big data's batch processing, agentic AI predominantly operates in real-time. The report highlights that 96 percent of all inference requests are processed instantaneously. This immediacy is particularly crucial in sectors where latency directly impacts value. The technology sector processes 32 real-time requests for every single batch request, while healthcare and life sciences, with applications like patient monitoring, show a 13-to-one ratio. This underscores the necessity for inference serving infrastructure capable of handling traffic surges without compromising user experience.
Governance Fuels Deployment
Perhaps a counter-intuitive finding for many executives is the symbiotic relationship between rigorous governance and deployment speed. Often perceived as a hindrance, robust governance and evaluation frameworks actually accelerate production deployments. Organizations employing AI governance tools launch over 12 times more AI projects into production compared to those that do not. Similarly, companies utilizing evaluation tools to systematically assess model quality achieve nearly six times more production deployments. Governance provides essential safeguards, such as defining data usage and setting rate limits, instilling confidence in stakeholders to approve deployments. Without these controls, pilot projects frequently remain trapped in the proof-of-concept phase due to unquantified safety or compliance risks.
Practical Automation: The 'Boring' Value
While autonomous agents often evoke futuristic visions, the current enterprise value derived from agentic AI lies in automating routine, essential, yet often mundane tasks. The primary AI use cases vary by industry but consistently address specific business challenges:
- Manufacturing and automotive: 35% of use cases focus on predictive maintenance.
- Health and life sciences: 23% of use cases involve medical literature synthesis.
- Retail and consumer goods: 14% of use cases are dedicated to market intelligence.
Moreover, 40 percent of leading AI use cases tackle practical customer issues, including support, advocacy, and onboarding. These applications deliver measurable efficiency gains and build the organizational capabilities required for more sophisticated agentic workflows.
Dael Williamson, EMEA CTO at Databricks, notes that the conversation for businesses has progressed from AI experimentation to operational reality. Williamson states that AI agents are already managing critical components of enterprise infrastructure, with organizations realizing true value by treating governance and evaluation as foundational elements rather than afterthoughts. He further emphasizes that competitive advantage is increasingly tied to how companies build, rather than merely what they acquire. Williamson suggests that open, interoperable platforms empower organizations to apply AI to their proprietary enterprise data, contrasting with embedded AI features that offer short-term productivity without long-term differentiation. For highly regulated markets, this blend of openness and control distinguishes successful pilots from competitive advantages.
This article is a rewritten summary based on publicly available reporting. For the original story, visit the source.
Source: AI News
