The State of Agentic AI, H2 2026: What the agents did wrong

A September 2026 survey of 1,030 people involved in AI systems found that agentic AI is increasingly deployed but accompanied by practical failures, security concerns, and demands for stronger oversight. Fifty-four percent of respondents reported an agent-related error in the prior 90 days, most commonly reversible workflow triggers, incorrect record updates, scheduling conflicts, and inappropriate communications, with incident rates rising as organizations operate more agents. While fully autonomous use grew from 4.2% to 7.5% since December, confidence in agents acting independently declined, and most teams use a middle-ground model in which agents can read data automatically but require approval to make changes. Auditability and identity management remain uneven: 53% can fully document an agent’s actions, while agents using dedicated identities are more easily audited than those acting through employee accounts or shared inboxes. Security approvals and integration complexity are now the leading barriers to implementation, and reliability and security controls outrank cost when organizations choose platforms. Although many respondents are considering replacing tools or vendors with AI agents, the report concludes that safe autonomy depends on clearly identifying whom an agent acts as, what it did, and which approval rules governed it.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.