AI-Harness is not affiliated with Harness Inc. (harness.io). AI-Harness does not provide DevOps, CI/CD, or software delivery tooling.

← Insights
Research digest

State of AI Agents 2026

This page summarizes publicly cited findings from Databricks’ State of AI Agents report (2026 edition). It is an independent digest for AI-Harness visitors; read the original for full methodology and charts.

Source: Databricks, State of AI Agents (2026). Report draws on aggregated platform telemetry and cited third-party surveys; dates and scope vary by chart.

Why this matters for operators

The report describes a shift from one-off chatbots to agentic workflows, heavier data and infrastructure demands, and a sharp split between pilots and production. Those themes align with how AI-Harness is built: measurable workflows, governance-minded execution, and human-led oversight—not generic model demos.

Key numbers (explicit callouts)

Reproduced from Databricks, State of AI Agents (2026), including cited third-party surveys and Databricks platform telemetry. Definitions, cohorts, and date ranges differ by chart—see the original PDF for full context.

Surveys & cited research

67%

Organizations already leveraging AI-powered tools (MIT Technology Review, cited in report).

>50%

Leaders who see agentic AI as a force multiplier for operations and decision-making (same cited survey).

19%

Organizations that have deployed AI agents—mostly to a limited extent (cited in report).

95%

Generative AI pilots that fail to reach production (2025 MIT NANDA report, cited in report).

40%

Respondents who believed their organization’s AI governance was insufficient (Economist Impact global survey, cited in report).

2%

Respondents rating their org’s AI performance highly on measurable business results (MIT Technology Review Insights, cited in report).

Databricks platform & product telemetry (as stated in report)

327%

Growth in multi-agent workflows in four months (report).

37%

Share of Agent Bricks usage: Supervisor Agent (October 2025).

31%

Share of Agent Bricks usage: Information Extraction (October 2025).

40%

Share of top AI use cases focused on customer experience & engagement (categorized use cases in report).

80%

Databases created by AI agents (Neon telemetry, October 2025 vs. 0.1% in October 2023).

97%

Database branches created by AI agents (same Neon telemetry; 0.1% two years prior).

50K+

Data and AI apps created since Databricks Apps public preview; 250% growth over six months (report).

78%

Companies using two or more LLM families (as of October 2025).

59%

Companies using three or more LLM families (October 2025 vs. 36% in July 2025).

83%

Retail companies using two or more model families (highest among industries called out).

96%

Share of requests processed as real-time vs. batch (Databricks platform, report).

32:1

Technology industry: real-time to batch request ratio (highest among industries).

13:1

Healthcare & Life Sciences: real-time to batch ratio (second highest).

Growth in AI Gateway usage (governance) since January 2025 (report).

12×

More AI projects put into production by companies actively using AI governance (report).

~6×

More AI projects into production for companies actively using evaluation tools (report).

Adoption and ambition

  • Cited MIT Technology Review research: 67% of organizations already use AI-powered tools; over half of leaders see agentic AI as a force multiplier—yet only 19% have deployed AI agents, mostly in a limited way.
  • Cited 2025 MIT NANDA: 95% of generative AI pilots fail to reach production—so “more pilots” alone is not a strategy; governance, evaluation, and operational fit matter.
  • Cited Economist Impact survey: 40% felt their AI governance program was insufficient for safety and compliance—parallel to governance and AI Gateway growth in the report’s own telemetry.

Multi-agent systems and use cases

The report cites 327% growth in multi-agent workflows in four months. On Databricks Agent Bricks (June–October 2025), Supervisor Agent reached 37% of usage by October 2025 (the top pattern), and Information Extraction was second at 31%.

Across categorized use cases, 40% of the top set relate to customer experience and engagement; the report stresses automating necessary but routine work (support, onboarding, marketing operations, compliance-adjacent tasks, etc.).

Infrastructure and the data layer

Neon telemetry (cited in the report): AI agents created 80% of databases by October 2025 (vs. 0.1% in October 2023), and 97% of database branches (vs. 0.1% two years prior)—illustrating how agents drive provisioning, testing, and experimentation at machine speed.

It also discusses demand for databases that support high-frequency, programmatic agent patterns—concurrency, branching, and real-time responsiveness—beyond traditional human-paced OLTP assumptions.

Models, real time, and ecosystem

As of October 2025, 78% of companies used two or more LLM families; 59% used three or more (up from 36% in July 2025). Retail led among industries at 83% using two or more families.

96% of requests on the Databricks platform were real-time vs. batch. The report highlights industry skew: Technology at 32 real-time requests per one batch request; Healthcare & Life Sciences at 13 to one.

Databricks Apps: 50K+ data and AI apps since public preview, with 250% growth over six months—evidence of “citizen” app creation at scale.

What actually gets AI to production

AI Gateway usage (governance) grew since January 2025. Companies actively using AI governance put 12× more AI projects into production; those using evaluation tools get nearly more projects into production—both quoted from the report’s analysis.

That mirrors how we talk about AI-Harness: workflows with clear ownership, measurable outcomes, and guardrails suitable for real operations—not unmanaged autonomy.

Next step

Turn insight into an execution plan

Book a session to map workflows, spend ceilings, and outcomes for your team—grounded in operations, not slide decks.