Databricks has just published “The State of AI Agents” summarizing its telemetry revealing that enterprise adoption of AI has spread well beyond copilots, isolated pilot projects, dashboards, and analysis functions, and is now widely entrusted with core systems.
“The State of AI Agents” specifies four key findings:
- Multi-agent systems are becoming the new enterprise operating model. Enterprises are transitioning from single chatbots to multiagent systems built on domain intelligence. Use of these systems grew by 327% in just four months.
- AI agents are driving core database activities. 80% of databases are built by AI agents. 97% of database testing and dev environments are now built by AI agents. This shift is driving the need for a new kind of database called Lakebase.AI is now part of critical workflows across industries. Most GenAI use cases are focused on automating routine necessary tasks, with 40% related to customer experiences. Model flexibility is the new AI strategy, with 78% of companies are using two or more LLM model families.
- AI evaluations and governance are the building blocks of production. Companies that use evaluation tools get nearly 6x more AI projects into production. Companies using AI governance put over 12x more AI projects into production. AI governance is a top investment priority, and grew 7x in nine months.
You can get the Databricks paper here: https://www.databricks.com/resources/ebook/state-of-ai-agents#:~:text=Key%20findings%3A&text=Enterprises%20are%20transitioning%20from%20single,more%20AI%20projects%20into%20production.
Sunil Gottumukkala, CEO, Averlon:
“When AI agents create databases at machine speed, ‘Secure by default’ becomes critical. Agents today optimize for the fastest path to completion, not safe configurations, so insecure defaults get replicated at scale. We saw this with row-level security gaps like the Moltbook incident. Teams need guardrails that catch risky configurations as they’re introduced and an operating model that prioritizes remediation when insecure defaults slip through.”
Ryan McCurdy, VP, Liquibase:
“When AI agents can create and modify database environments on demand, the database becomes a high frequency software event. The risk is uncontrolled change. Policy enforced in the workflow, automatic audit evidence, drift detection, and trusted rollback are essential to keep velocity without sacrificing control.
“Moreover, agentic development will multiply database changes. If governance stays manual, you get drift, surprise outages, and you can’t explain what changed when it matters. Database Change Governance is how enterprises keep the data layer fast, trusted, and auditable as it goes agentic.
“The answer isn’t more humans reviewing more changes. It’s policy enforced in the workflow, automatic evidence capture, and trustworthy rollback.”
John Carberry, Solution Sleuth, Xcape, Inc.
“The discovery that 80% of new enterprise databases are currently created by AI agents signifies a historic transition from human-centric administration to “vibe coding” on an industrial scale. Although this increase in autonomous infrastructure speeds up development, it also adds a significant “governance debt” by directly incorporating security logic into AI-generated code that is rarely submitted to human peer review.
“The main risk is “excessive agency,” whereby these agents might unintentionally produce vulnerable endpoints, excessively lenient access rules, or unsafe schemas that get beyond conventional perimeter defenses. Moreover, these databases produce a vast, undetectable attack surface called Shadow Data, which is usually left out of centralized logging and auditing because they are routinely spun up in real-time “branches” for testing and development. In response, SOC teams must switch from post-deployment scanning to infrastructure-level enforcement, in which the security border is located outside of the code that is generated and checks each database operation against a policy that is hardcoded at runtime. The function of the DBA is changing from being a builder to a high-level auditor of autonomous systems as AI progresses beyond creating chatbots to designing the enterprise’s basic foundations.
“The ‘human in the loop’ becomes a myth when 80% of your data infrastructure is built by AI.”
AI Agents Now Building 80% Of Certain Key Enterprise Infrastructure – data & cyber experts comment
Posted in Commentary with tags Databricks on February 6, 2026 by itnerdDatabricks has just published “The State of AI Agents” summarizing its telemetry revealing that enterprise adoption of AI has spread well beyond copilots, isolated pilot projects, dashboards, and analysis functions, and is now widely entrusted with core systems.
“The State of AI Agents” specifies four key findings:
You can get the Databricks paper here: https://www.databricks.com/resources/ebook/state-of-ai-agents#:~:text=Key%20findings%3A&text=Enterprises%20are%20transitioning%20from%20single,more%20AI%20projects%20into%20production.
Sunil Gottumukkala, CEO, Averlon:
“When AI agents create databases at machine speed, ‘Secure by default’ becomes critical. Agents today optimize for the fastest path to completion, not safe configurations, so insecure defaults get replicated at scale. We saw this with row-level security gaps like the Moltbook incident. Teams need guardrails that catch risky configurations as they’re introduced and an operating model that prioritizes remediation when insecure defaults slip through.”
Ryan McCurdy, VP, Liquibase:
“When AI agents can create and modify database environments on demand, the database becomes a high frequency software event. The risk is uncontrolled change. Policy enforced in the workflow, automatic audit evidence, drift detection, and trusted rollback are essential to keep velocity without sacrificing control.
“Moreover, agentic development will multiply database changes. If governance stays manual, you get drift, surprise outages, and you can’t explain what changed when it matters. Database Change Governance is how enterprises keep the data layer fast, trusted, and auditable as it goes agentic.
“The answer isn’t more humans reviewing more changes. It’s policy enforced in the workflow, automatic evidence capture, and trustworthy rollback.”
John Carberry, Solution Sleuth, Xcape, Inc.
“The discovery that 80% of new enterprise databases are currently created by AI agents signifies a historic transition from human-centric administration to “vibe coding” on an industrial scale. Although this increase in autonomous infrastructure speeds up development, it also adds a significant “governance debt” by directly incorporating security logic into AI-generated code that is rarely submitted to human peer review.
“The main risk is “excessive agency,” whereby these agents might unintentionally produce vulnerable endpoints, excessively lenient access rules, or unsafe schemas that get beyond conventional perimeter defenses. Moreover, these databases produce a vast, undetectable attack surface called Shadow Data, which is usually left out of centralized logging and auditing because they are routinely spun up in real-time “branches” for testing and development. In response, SOC teams must switch from post-deployment scanning to infrastructure-level enforcement, in which the security border is located outside of the code that is generated and checks each database operation against a policy that is hardcoded at runtime. The function of the DBA is changing from being a builder to a high-level auditor of autonomous systems as AI progresses beyond creating chatbots to designing the enterprise’s basic foundations.
“The ‘human in the loop’ becomes a myth when 80% of your data infrastructure is built by AI.”
Leave a comment »