According to Databricks, global usage of AI agents has increased by 327% in just four months, positioning it as a key new feature in AI adoption. This scenario represents significant progress on the path of agent AI, from theoretical models to integration backed by measurable business outcomes.
The move from traditional large-scale language models (LLMs) to multi-agent systems addresses the need to automate complete, specialized workflows rather than individual tasks. According to the 2026 State of AI Agents report, this rapid growth is attributed to multi-agent architectures that form “new business operating models.”
An example of one of these scenarios is an organization integrating “Supervisor Agent,” a Databricks solution that coordinates different agents and tools to perform highly complex processes. This allows the system to not only reason, but also independently plan and execute actions based on the specific context of the enterprise. Since its release in July 2025, Supervisor Agent usage has become a leading use case, accounting for 37% of total activity on the Databricks platform by October 2025.
Since the large-scale emergence of generative AI (GenAI) three years ago, the industry has shifted from an exploratory approach to one focused on key operational outcomes. Today, 66% of organizations are already using AI-powered tools, but only 19% have implemented autonomous agents, indicating significant scope for expansion in the coming years.
The current relevance of AI agents lies in their ability to bridge the gap between technical possibilities and business value. According to a Databricks report, this phenomenon comes at a time of “once-in-a-decade” infrastructure change for the data layer. The report also agrees that leading companies are moving away from pursuing a “single model” to adopting a multi-model flexibility strategy, with 78% of companies using two or more model families to optimize performance and cost depending on the task.
Data layer automation and “Vibe coding”
One of the report’s most shocking findings is the autonomy agents have achieved in managing their infrastructure. Telemetry data shows:
-
80% of current databases are created by AI agents, a significant increase from 0.1% in 2023.
-
97% of test and development environments (database branches) are built by agents, reducing provisioning time from hours to seconds.
This progress is being driven by the emergence of “vibe coding,” where users write requirements in natural language for AI to generate corresponding code. Gartner estimates that by 2028, 40% of new product software will be created using these technologies. To support this load, a new generation of operational databases, such as Lakebase, have emerged that are designed to handle the high read and write frequencies generated by autonomous agents.
Ratings and distribution by industry
The report also identifies a direct correlation between management tool use and project success. The market responded with a seven-fold increase in investment in AI governance over nine months.
Efficiency metrics are compelling. Companies with unified governance protocols put 12 times more AI projects into production than those without. Similarly, systematic evaluation tools (customized benchmarks) enable nearly six times as many projects to be put into production, ensuring accuracy and security of response.
The technology sector is leading the way in adoption, building nearly four times as many multi-agent systems as any other industry. However, the applications of AI are diversifying, including:
-
The retail industry is the area where multi-model use is most likely to occur, with 83% employing two or more LLM families.
-
Customer experience accounts for 40% of global use cases, covering technical support, onboarding, and personalized marketing.
-
Latin America shows a pragmatic approach with loan origination as the main use case (10%). Additionally, 77% of inference requests in the region are processed in real-time, highlighting the importance of low latency in emerging markets.
The report concludes that the future of enterprise AI will move away from opting for decoupled models and shift toward a focus on integrating business context and autonomous execution tools. Multi-agent systems are expected to evolve towards a continuous learning model, where real-time evaluation allows agents to adapt to changes in the environment without continuous human intervention.
