Databricks: Only 19% of Organizations Have Deployed AI Agents. But They’re Already Creating 97% of Databases.

10 Interesting Learnings from Databricks’ State of AI Agents 2026

Databricks recently released their State of AI Agents report, built from data across 20,000+ organizations including over 60% of the Fortune 500. The numbers are worth sitting with. Here are the 10 that matter most for B2B founders and operators.

#1. Only 19% of organizations have actually deployed AI agents

The big takeway — for right now. Everyone is talking about agents. Most haven’t deployed one yet. 67% of organizations are using AI-powered tools in some form, but actual agent deployment is still the minority position. That gap between talking about agents and running them in production is the whole story of 2025 and probably most of 2026.

#2. Multi-agent systems grew 327% in just four months

The shift from single-purpose chatbots to coordinated, multi-agent systems is not slow. On the Databricks platform, multi-agent workflow usage grew 327% in four months. Once organizations figure out how to orchestrate agents across specialized domains, adoption accelerates fast. This is not a gradual adoption curve. It’s a step function.

#3. AI agents now build 80% of databases. Two years ago it was 0.1%

This is one of the most striking data points in the entire report. In October 2023, 0.1% of databases on Neon (the serverless Postgres layer Databricks acquired) were created by AI agents. By October 2025, that number was 80%. That is not gradual progress. That is a complete handoff of a core infrastructure function to agents in 24 months.

#4. 97% of database branches are now created by AI agents

Same story, even more pronounced. Database branches, the isolated environments used for testing and development, went from 0.1% agent-created to 97% in two years. Human engineers are largely out of this loop now. The operational implications for database infrastructure vendors are enormous.

#5. Tech companies build nearly 4x more multi-agent systems than any other industry

Across all industries tracked, technology companies deploy Supervisor Agent architectures at nearly four times the rate of the next closest vertical. The reason makes sense: tech teams are better at decomposing complex business problems into discrete, solvable sub-problems that agents can handle in parallel. Other industries will get there, but the tech sector is currently running a different race.

#6. 78% of companies are now running two or more LLM model families

The “pick one model and commit” era is over. As of October 2025, 78% of companies on the Databricks platform use two or more LLM families, whether that’s GPT, Claude, Llama, Gemini, Qwen, or others. More telling: the share using three or more model families jumped from 36% to 59% in just three months, between July and October 2025. Multi-model is the default enterprise AI strategy now.

#7. 40% of the top AI use cases are customer experience work

Despite all the internal productivity talk, the biggest concentration of enterprise AI deployments focuses on customers: support, onboarding, advocacy, personalized content. That’s where the ROI case is clearest and the volume of repetitive work is highest. Healthcare focuses on medical literature synthesis. Manufacturing and energy focus on predictive maintenance. But across all industries, customer-facing work dominates.

#8. Over 50,000 data and AI apps were built on the Databricks platform in roughly a year, growing 250% in the last six months of that period

Vibe coding, where non-engineers describe what they want and AI writes the code, has moved from concept to measurable production phenomenon. The citizen developer is real and the numbers are accelerating. What’s interesting about the Databricks data is the workflow it reveals: business users build the prototype through vibe coding, then data teams and engineers take what actually works and productionize it for enterprise deployment. That’s a fundamentally different software development motion than anything that existed three years ago. Gartner estimates 40% of new production software will be built this way by 2028. Based on what Databricks is seeing in their platform data, that estimate may be conservative.

#9. Companies using AI governance put 12x more projects into production

This is the clearest proof point in the report. Organizations that implement unified AI governance, defined policies around how data is used, guardrails, rate limits, structured accountability, ship twelve times more AI projects than those that don’t. AI governance usage on the Databricks platform grew 7x in nine months. The enterprises that treat governance as an accelerator rather than a compliance checkbox are the ones actually deploying at scale.

#10. Companies using AI evaluation tools get 6x more projects into production

Governance and evaluation work together. Evaluation tools that continuously test accuracy, safety, and compliance against an enterprise’s own data and KPIs, not generic benchmarks, are what close the pilot-to-production gap. Organizations using these tools get six times more AI projects into production than those that don’t. Evaluation is the mechanism that closes that gap.

A few more data points worth noting:

96% of all AI inference requests are served in real time, not batch. Subsecond AI is the expectation now across every industry. Supervisor Agent went from zero to the #1 agent use case within four months of its July 2025 launch, capturing 37% of all Agent Bricks usage by October. When the right orchestration primitive exists, enterprises adopt it fast. Retail leads all industries in multi-model AI adoption, with 83% of companies running two or more LLM families, deliberately mixing models to balance performance against cost by task. The Tech industry builds nearly 4x more multi-agent systems than any other sector and processes 32 real-time AI requests for every single batch request. Both numbers point to the same thing: tech companies have operationalized agents faster and more deeply than anyone else.

Source: Databricks State of AI Agents 2026, based on aggregated, anonymized data from 20,000+ organizations including 60%+ of the Fortune 500.