Sinch finds 74% rolled back agents

Sinch said on May 13 that 74% of enterprises have already rolled back or shut down an AI customer-communications agent after deployment, citing governance failures in a new report called *The AI Production Paradox*. The company said the findings came from an independent survey of 2,527 senior decision-makers across 10 countries and six industries. The same study said 62% of enterprises already have AI agents live in production in customer communications, suggesting the problem is not a lack of pilots but trouble operating systems after launch. ### Why are companies pulling agents back after they were already live? Sinch said the rollback figure reflects “governance failure” after deployment rather than hesitation before launch. The company said enterprises are running into operational problems once AI agents reach customer-facing use, even as adoption has moved into production environments. (sinch.com) The Fast Mode, citing the Sinch report, said companies reported poor outcomes and operational issues after deployment. Sinch’s own release did not frame the problem as model capability alone; it tied the failures to production controls and guardrails. (sinch.com) ### Why does the 81% figure matter more than the headline 74%? Sinch said the rollback rate rises to 81% among organizations with fully mature guardrails. That suggests, at minimum, that even companies further along on governance are still encountering failures once live customer agents are exposed to real workflows and real users. That last point is an inference from the survey breakdown, not a direct company statement. (thefastmode.com) The study’s sample size—2,527 senior decision-makers across 10 countries and six industries—gives the figure breadth, though the data still comes from a vendor-backed report. Sinch presented the results as evidence of a “production paradox”: enterprises are deploying AI agents, but many are also retreating from them after launch. (sinch.com) ### If full autonomy is stumbling, what are buyers backing instead? Observe.AI said on May 19 that Metrigy named it a 2026 MetriStar Top Provider for agent-assist applications. The company said the recognition was based on business impact and customer sentiment for AI-powered agent-assist tools. (sinch.com) Metrigy said its 2026 CX MetriStar program recognized 74 winners across nine categories, including both agent assist and AI agents. The firm said its awards draw on customer sentiment ratings and reported improvements in metrics such as customer satisfaction, employee efficiency, revenue and operating costs. (prnewswire.com) ### What does “agent assist” mean in practice? Observe.AI said its Companion Agent platform provides real-time support, guidance and automation for frontline customer-service teams. That keeps a human agent in the loop while software handles prompts, recommendations and workflow support. (metrigy.com) Metrigy’s separate award categories for agent assist and AI agents point to a distinction buyers are making in the market: tools that help human agents versus tools that act on their own. The Sinch findings and the Observe.AI recognition together indicate that vendors and enterprise buyers are still investing in AI for customer service, but with tighter operating boundaries. That comparison is an inference based on the two announcements and Metrigy’s category structure. (observe.ai) ### What should readers watch next? Metrigy’s 2026 CX MetriStar winners page lists agent assist and AI agents as separate tracked categories, and Sinch’s *AI Production Paradox* report is now the main source for the rollback data. Any next signal from the market is likely to come from additional vendor disclosures, customer case studies or updated survey data showing whether rollback rates fall as controls improve. (metrigy.com)