Databricks reports surge in multi-agent AI systems transforming enterprise workflows

According to Databricks’ 2026 State of AI Agents report, multi-agent systems are accelerating across industries, reshaping data infrastructure, and demanding new governance practices as enterprises shift from proof-of-concept to deployment.

Organisations are moving past proof‑of‑concepts and simple chatbots toward agentic systems that plan, reason and act across real business processes, according to Databricks’ 2026 State of AI Agents report. The company’s ana...

lysis of aggregated, anonymised telemetry from more than 20,000 customers shows a sharp acceleration in multi‑agent adoption and a parallel shift in how enterprises design, evaluate and govern AI.

Databricks found multi‑agent workflows on its platform grew 327% between June and October 2025, with technology firms building such systems at nearly four times the rate of other industries. Across sectors, leaders are pairing multiple model families to match capabilities to tasks: in October 2025, 78% of Databricks customers used two or more model families, while the share using three or more rose from 36% in August to 59% in October. Retail leads cross‑model use, with 83% of firms in that sector deploying two or more models.

The report describes a pragmatic, sector‑focused wave of deployments. Industry patterns include predictive maintenance in manufacturing and automotive (35% of use cases), medical literature review in health and life sciences (23%), and market insight and research in retail and consumer goods (14%). Overall, 40% of the principal AI use cases observed were directed at practical customer problems such as support and onboarding, and 96% of observed AI requests were processed in real time to enable copilots, personalisation engines and interactive assistants.

A major structural change is underway in data infrastructure as agents become active participants rather than mere application layers. Databricks reports AI agents now create 80% of databases on its platform and build 97% of test and development branches, a trend the company links to so‑called “vibe coding,” where non‑specialist business users express intent while agents perform routine implementation. Since the Public Preview of Databricks Apps, the firm says more than 50,000 data and AI apps have been created, with usage up 250% over six months.

Databricks has been productising agent design with tools such as Agent Bricks, introduced in June 2025, which it says automates optimisation of agents against customers’ own data for common industry tasks ranging from structured extraction to orchestrated multi‑agent systems. According to Databricks’ blog, Supervisor Agents, those that coordinate multiple specialised agents, account for the largest share of agent activity on Agent Bricks.

The transition is not without friction. External research cited in the report shows 95% of generative AI pilots in 2025 never reached production, a gap Databricks and industry commentators attribute to attempts to apply traditional software development practices to probabilistic, agentic AI. Dael Williamson, CTO at Databricks EMEA, argues top performers focus less on exhaustive upfront design and more on robust testing, evaluation and guardrails; the study Databricks cites finds firms applying active AI governance put twelve times as many projects into production, while those using formal evaluation frameworks deliver six times as many production deployments.

Industry observers see broader implications for platform leadership. According to Forbes, Databricks ranked third in enterprise data warehousing adoption as of February 2026 with roughly 15% penetration, behind Snowflake at 62% and Amazon Redshift at 29%. However, Forbes notes Databricks has differentiated through generative AI capabilities and unified analytics and is positioning its Lakebase technology as a foundation for a new class of “active” databases. “The database is the system of record for AI applications. It’s no longer just a place to store rows; it’s the persistent memory and coordination layer for multi‑agent systems,” said Alexey Shamgunov to Forbes, describing Lakebase as an attempt to make the database part of the fabric that synchronises agent logic with enterprise data.

Market forecasts point to rapid further change. Gartner estimates that by 2028, 90% of enterprise software engineers will use AI code assistants, up from below 14% in early 2024, reinforcing the view that developer workflows and software lifecycles will be reshaped as agents take on more of the routine coding and environment provisioning.

The growing emphasis on governance and evaluation reflects the new risk landscape. Databricks and other commentators highlight concerns about “agent drift,” security exposures and brittleness as underlying data, APIs or business rules change. For organisations that have bridged the production gap, Databricks and trade coverage report transformative returns: agents embedded in critical workflows drive faster experimentation, shorter time to deploy and new modes of intent‑based computing in which users specify desired outcomes rather than step‑by‑step instructions.

As the enterprise AI stack evolves, competition will centre on platforms that combine multi‑model support, scaleable data fabrics and integrated governance. Databricks’ findings portray agentic systems as entering the mainstream, but they also underline that converting prototypes into reliable, governed production services remains the key operational challenge.

Source: Noah Wire Services