The pitch is everywhere now. AI agents that book your travel, manage your calendar, handle your email, coordinate your projects. Autonomous systems that take a goal and figure out how to accomplish it. Tell the agent what you want. It handles the rest.
The demos are impressive. I’ve watched agents navigate complex multi-step workflows, make decisions, recover from errors. The technology clearly works.
And yet.
Talk to anyone actually deploying these systems in production and you hear a different story. Impressive in controlled conditions. Brittle in the real world. Requires more human supervision than the marketing suggested. Works great until it doesn’t, and when it doesn’t, it fails in ways that are hard to predict and harder to fix.