Serverworks: 81.1% of IT engineers say AI agents judge better than humans

The survey, run by Serverworks between June 29 and July 5, 2026 and published on August 20, captures a workforce that has already moved past awareness into use. On the personal side, 61.1% of the 260 engineers said they use AI agents, with 28.8% using them daily. At work, 52.7% reported some form of adoption. This is not a picture of hesitation about the technology itself.

The hesitation appears when the question turns to responsibility. Respondents were most eager for agents to take over simple-task automation (25.4%) and to raise development speed (22.7%), and they were least willing to grant full autonomy for incident response and production deployment. Their top concerns — misjudgment at 16.5%, information leakage at 14.6%, and hallucination at 14.2% — point to consequences rather than capability. That reading is consistent with the approval split: 44.2% said human approval is needed for many tasks but not all, and 26.2% said it is always needed.

The stakes therefore seem to hinge less on whether AI agents can judge well than on where organizations draw the approval line. The 73.1% who expect some autonomous task execution by 2030, and the 73.4% who say they want to actively use agents, suggest the direction of travel is toward more delegation. For engineering managers and operations leads, the near-term test is likely whether approval rules, rather than model quality, determine how much of that delegation actually happens.