Something Has Gone Terribly Wrong with AI Agents

The pitch was autonomous. The demos were polished. Ninety-seven percent of companies deployed something they called an agent. Eleven percent are running one in production at scale. The gap between those numbers is where the truth lives.

Created by Gemini

The pitch was always clean. Deploy an agent. Give it your tools. Watch it work while you sleep. Autonomous pipelines handling research, filing, scheduling, customer support entire workflows delegated to a system that doesn’t need to be reminded, doesn’t get tired, doesn’t ask for clarification. Every major lab shipped an agent framework. Every major consultancy published a report about the agentic future. Every investor deck for eighteen months has included the phrase “agentic AI” somewhere in the first three slides.

Now the data is in. And it tells a story that none of those decks bothered to include.

In a 2026 survey of 2,400 enterprises conducted by WRITER and Workplace Intelligence, 97 percent of companies said they had deployed AI agents in some form over the previous twelve months. McKinsey’s 2026 data found that only 11 percent of those organizations were running an agent in production at genuine scale not a sandboxed pilot, not a demo environment, not a project with one user and a sympathetic product manager. Genuinely in production. Genuinely doing real work.