A new global study from Sinch AB challenges the assumption that getting AI agents into production is the hardest part. The real problem, the data suggests, is keeping them there.
The report, The AI Production Paradox, surveyed 2,527 senior decision makers across 10 countries and six industries. It found that 62% of enterprises already have AI agents live in customer communications. Yet 74% have rolled back or shut down at least one deployed agent after launch. Among organizations with the most mature governance frameworks, that rollback rate climbs to 81%.
Why rollbacks happen
The primary cause cited is governance failure. 62% of respondents pointed to data quality concerns, and 45% flagged regulatory compliance as a major hurdle. But the report's key finding is that better governance does not correlate with lower rollback rates — it correlates with higher ones. Organizations with mature guardrails are not failing more; they are detecting failures earlier, thanks to better monitoring.
Sinch CPO Daniel Morris described the dynamic as a "guardrail tax": engineering teams spend most of their time building and maintaining safety systems instead of improving the customer experience. 84% of AI engineering teams report spending at least half their time on safety infrastructure.
Investment priorities shift
Enterprises are investing more in trust, security, and compliance (76%) than in AI development itself (63%). That makes governance the single largest investment category in AI programs. Yet the data shows that governance investment alone is not solving the problem.
Infrastructure as the differentiator
The strongest predictor of successful AI deployment, according to the study, is satisfaction with communications infrastructure — stronger than both investment levels and guardrail maturity. 87% of organizations rate high-performance infrastructure as essential or very important. Still, most say their current provider falls short in at least one meaningful area. 55% of enterprises are building custom infrastructure to manage cross-channel context, and 86% have evaluated or are actively evaluating new communications providers.
Outlook
Despite the high rollback rates, 98% of enterprises report increasing AI investment in 2026. The challenge has shifted from deployment to maintaining performance, reliability, and control once AI is live. The report suggests that the industry's focus on governance as a cure-all may be misplaced — infrastructure quality appears to matter more.
Bottom line
If your organization is deploying AI customer agents, expect to invest heavily in monitoring and infrastructure. High rollback rates are not necessarily a sign of failure; they may indicate that your detection systems are working. The real risk is spending so much on safety scaffolding that you have no resources left for improving the actual customer experience.