Temporal survey exposes AI agent proof gap

Agent adoption is accelerating faster than engineering teams can operationalize it. The real bottleneck is no longer model access—it is building systems that preserve state, surface failures, and earn trust in production.

• –Daily-or-more agent use reached 80.8%, up from 47.3% a year earlier; respondents now run a median of five agents.
• –Tracking state ranked as the top blocker at 35.7%, followed by debugging and token or compute costs.
• –Trust remains conditional: only 24.7% completely trust agent outputs, while 41.1% still encounter issues daily or more.
• –39.5% cite security concerns as the biggest barrier to fully autonomous agents, reinforcing the need for human review and strong guardrails.
• –The findings are self-reported and commissioned by Temporal, so the emphasis on state and reliability should be read alongside the company’s durable-execution business.