Today I read 74% of companies pulled their AI agents after deploying them. obviously we don't hear about this from the news

today i read that 74% of companies that deployed AI agents in production pulled them back. not pilots. not tests. live systems. shut down. the study surveyed 2,500+ senior decision makers across 10 countries. these weren't small experiments most of these orgs had dedicated AI teams and real budgets behind them. but here's the part that actually got me: The rollback rate was 81% among companies with the MOST mature AI governance. the ones doing everything right were failing more visibly than the ones who were winging it. turns out better governance doesn't prevent failures. It just means you catch them faster and have a process to shut things down. which is actually the goal but nobody frames it that way. I think what was actually breaking is not the models. It’s the infrastructure underneath them. agents were hallucinating in production because they had no clean, structured context to pull from. old databases. disconnected tools. data living in spreadsheets someone emails on fridays. the AI wasn't the problem. everything the AI was supposed to work with was the problem. i've seen this firsthand building agents for our own ops. the ones that stuck weren't the most technically impressive. they were the ones connected to clean, organized context. the ones we abandoned fast were the ones we pointed at messy, inconsistent information and hoped for the best. The article's conclusion was: "the next wave of enterprise AI won't be defined by which model you're running. it'll be defined by whether your foundation can support it." Any thoughts on why AI agents are failing in production