Gartner projects that 40% of agentic AI projects will be canceled by end of 2027. Not because the models weren't capable, but because teams deployed without the governance controls to catch errors before they compounded. Forrester puts it more directly: an agentic AI deployment will cause a major enterprise breach this year. The conditions already exist in most organizations.

Meanwhile, SREs and platform engineers are absorbing a different kind of pressure. Agents are generating more code, more PRs, more decisions — but the humans reviewing them haven't scaled with the volume.

At DevOps Experience 2026 on September 24, leading engineers and architects who are working through these problems in production will break down how to scope agent access before governance becomes a crisis, where human oversight has to stay as autonomy expands, and how incident response changes when the agent is what caused the incident.