State of AI Agents 2026

Published by Databricks on September 8, 2026 · Updated September 8, 2026

• ai-agents
• enterprise-ai
• database-infrastructure
• llm-governance

What does the State of AI Agents report say about enterprise AI adoption?

Enterprise AI is moving from pilots and chatbots to agentic architectures that combine foundation models, enterprise context, and tools so systems can plan and take action. Databricks says the organizations succeeding are aligning AI use cases to business goals, building focused agents for specific operational tasks, and updating their data layer for a shift that happens only once every decade.

• MIT Technology Review found 67% of organizations are already using AI-powered tools.
• More than half of leaders see agentic AI as a force multiplier for operational performance and decision-making.
• Only 19% of organizations have deployed AI agents, and most deployments are still limited.

How did Databricks create this report?

The 2026 State of AI Agents is built from fully aggregated, anonymized data collected from Databricks customers using the Databricks Data Intelligence Platform. It includes more than 20,000 organizations worldwide, over 60% of the Fortune 500, and data from every major industry and company size, from startups to large enterprises.

Whenever possible, the report analyzes data from November 1, 2024 to October 31, 2025. Because AI products launch quickly and a full year of data is not always available, some charts use shorter date ranges and are labeled accordingly. Usage is measured by the number of active, engaged customers, and minimum bars are used when needed to represent growth accurately.

What is the difference between a chatbot pilot and a real enterprise AI agent?

A real enterprise AI agent does more than answer prompts. It reasons, coordinates tools, and orchestrates workflows across specialized tasks and enterprise data.

Chatbots were an early foundation for enterprise AI and remain a common use case, but Databricks says the next phase is multi-agent systems that can autonomously plan and execute complex work. These systems are built on domain intelligence rather than generic conversation alone.

A financial services organization, for example, may combine intent detection, document retrieval, and compliance checks to deliver personalized responses to clients.

How are organizations moving from experimentation to production-ready AI agents?

Organizations are moving to production-ready AI agents by focusing on specific business outcomes instead of broad experimentation. Databricks says successful enterprises are aligning AI use cases to business goals, developing focused agents, and using multiple models and tools with their own data.

The report shows that 78% of companies are using two or more LLM model families, which makes model flexibility the new AI strategy. Databricks also reports that AI evaluations and AI governance are key production enablers: companies that use evaluation tools get nearly 6x more AI projects into production, and companies that use AI governance put over 12x more AI projects into production.

Governance is also rising quickly as an investment priority, growing 7x in nine months.

Which AI agent use cases are growing fastest in Databricks Agent Bricks?

Supervisor Agent and Information Extraction are leading the current Databricks Agent Bricks usage mix. Since Databricks launched Supervisor Agent in July 2025, it has quickly become the No. 1 agent use case and accounted for 37% of all Agent Bricks usage in October 2025. Information Extraction is the second most commonly used agent and represents 31% of usage.

Databricks Agent Bricks includes four agent use cases:

• Custom LLMs: generate text-based results for domain-specific tasks.
• Supervisor Agent: orchestrates agents and tools to execute complex tasks.
• Knowledge Assistant: custom-trains Q&A chatbots on enterprise documents.
• Information Extraction: transforms unlabeled text documents into structured tables.

These patterns show that organizations are using agents to coordinate work, surface knowledge, and turn unstructured information into usable data.

What kinds of workflows are best suited for agentic systems?

Agentic systems are best suited for routine, necessary business workflows that benefit from reasoning, coordination, and access to proprietary data. Databricks says 40% of AI use cases focus on customer experiences and engagement, including customer support, customer advocacy, onboarding, and personalized marketing content.

The report’s categorized AI use cases also include market intelligence, predictive maintenance, claim processing, anti-money laundering, medical literature synthesis, order process automation, regulatory reporting requirements, loan origination, customer interaction summarization, clinical note summarization, and customer sentiment analysis from support logs.

Across industries and regions, the top use cases are practical and operational. For example, Healthcare and Life Sciences emphasizes medical literature synthesis, while Manufacturing and Automotive and Energy and Utilities prioritize predictive maintenance.

What infrastructure changes do companies need before they can run AI agents at scale?

AI agents require a database and infrastructure shift because they create, branch, and query systems at a pace traditional OLTP designs were not built to handle.

Databricks says AI agents now create 80% of databases and 97% of database testing and development environments, based on telemetry from Neon, the serverless Postgres database acquired by Databricks and the core technology behind Databricks Lakebase. Branches are isolated, version-controlled database variants that agents can create quickly for testing and experimentation.

The report says this shift is driving the need for a new kind of database called Lakebase. Traditional OLTP databases were built around human workflows, predictable transactions, infrequent schema changes, and manual provisioning. Agentic workflows are different: they involve continuous read and write patterns, programmatic creation and teardown of environments, and complex read-intensive reasoning loops.

Why are governance and evaluation critical for production AI agents?

Governance and evaluation are critical because they make enterprise AI outputs more accurate, more reliable, and easier to move into production. Databricks says AI evaluations are essential to producing high-quality outputs, and companies that use evaluation tools get nearly 6x more AI projects into production.

AI governance has an even larger production effect: companies that use AI governance put over 12x more AI projects into production, and governance investment grew 7x in nine months.

For technical leaders, the production test is whether an AI agent can consistently produce accurate results on enterprise data, operate safely inside governed workflows, and complete the full multi-step task without breaking orchestration.