Databricks' 2026 State of AI Agents report, based on aggregated and anonymized activity from more than 20,000 Databricks customers worldwide, found that AI agents now create roughly 80% of new databases and 97% of test and development environments provisioned on the platform, a striking shift in who, or what, is doing the foundational infrastructure work that used to be entirely manual data engineering labor. The report also found that multi-agent workflows, setups where multiple specialized AI agents coordinate on a task rather than a single agent working alone, grew 327% over just four months, driven by the rapid introduction of agent orchestration features across the data tooling ecosystem. But the report's most important finding for data engineering leaders is the gap it surfaces between experimentation and production maturity: despite this explosive growth in agent-created infrastructure, only 19% of audited organizations report having deployed AI agents at genuine scale, meaning the vast majority of this agent-driven database and environment creation is still happening in exploratory, non-production contexts rather than being trusted with mission-critical systems. That combination, agents already doing the bulk of routine infrastructure provisioning work while still largely excluded from production-critical decisions, matches a broader pattern showing up across software engineering more generally: the manual, repetitive tasks that used to train junior engineers, standing up a test database, provisioning a dev environment, are being automated away first, while judgment-heavy work, deciding what actually belongs in production, remains a human responsibility for now. For data engineering leaders, the report is a useful benchmark for calibrating internal AI-agent adoption against industry norms, and a reminder that provisioning automation and production trust are two very different maturity milestones.