Already 4 months old when the scout caught it.
Agent sprawl
What it is
The uncontrolled proliferation of AI agents across an organization, deployed piecemeal by individual teams without a shared inventory, ownership model, or governance layer. A developer automates one workflow with an agent, it works, other teams copy the pattern, and within months an org has dozens to thousands of agents touching production systems — each with its own permissions, credentials, and LLM provider, and no central view of what any of them can do.
Why it's emerging now
2026 is the year agent deployment went from experimental to default at many companies, and the numbers caught up with the risk: surveys this year put 96% of enterprises running agents in production, with the large majority of respondents admitting sprawl is already out of hand and only a small fraction able to fully inventory their own agents. Gartner has been publishing governance guidance specifically framed around this term through the spring and summer, and SAP put it on its news site as a board-level issue in August. It's the multi-agent-era successor to "shadow IT" — the infrastructure got easy enough to stand up that governance couldn't keep pace.
How it relates to what you know
It's a downstream consequence of multi-agent orchestration and agent harnesses (both on the ledger) becoming cheap and common: the same ease of spinning up an agent that makes orchestration powerful also makes uncontrolled proliferation possible. It sits closer to AI governance and security than to a modeling technique — the practitioner conversation around it is about agent registries, permission scoping, and audit trails, not model architecture.