There’s a confidence gap running through the enterprise AI boom, and Sinch just put a number on it. In a global survey of 2,527 senior decision-makers, the Swedish cloud communications company found that 60% of C-suite executives say they are very confident in their organization’s AI programs — but only 43% of the directors and managers actually building and running those systems agree.
Sinch calls the report “The AI Production Paradox,” and the framing fits. AI is clearly in production: 62% of enterprises say they already have AI agents live. The problem is keeping them there. A striking 74% of organizations report having rolled back or shut down a deployed AI agent — a figure that climbs to 81% among companies with mature governance frameworks, which suggests the teams watching most closely are also the quickest to reach for the kill switch.
Guardrails are eating engineering
The study, run with an independent research institute between January and February 2026, spanned ten countries including the US, UK, India, Germany and Brazil, across finance, healthcare, telecom, technology and retail. Its most telling stat may be where the effort goes: 84% of AI engineering teams say they spend at least half their time on guardrails rather than building. And 75% report investing more in trust, security and compliance than in AI development itself (63%) — a sign that the hard part of shipping AI isn’t the model, it’s everything around it.
That “everything around it” is, conveniently, Sinch’s pitch. The company found that satisfaction with communications infrastructure was the single strongest predictor of confidence in AI deployment, outranking governance maturity, deployment experience and even investment levels. Read charitably, it’s a genuine finding; read cynically, a company that sells communications infrastructure has produced research concluding that communications infrastructure is what’s missing.
“For years, the AI conversation has largely focused on whether organizations can get AI into production. Our research suggests another challenge is emerging: making sure leadership and operational teams are working from the same reality.”
Sophie Cheng, Chief Marketing Officer, Sinch
The churn underneath is worth watching. 86% of respondents said they have evaluated or are considering new communications providers, and 98% plan to increase AI spending in 2026 — a market that is simultaneously dissatisfied and unwilling to slow down. For all the rollbacks, nobody is turning off the tap. Another 55% say they have had to build custom infrastructure just to carry context across channels, the kind of plumbing work that rarely makes the roadmap but quietly determines whether a chatbot remembers what you told it thirty seconds ago.
The bigger signal here is cultural, not technical. Executives are tracking deployments and budgets; the people closer to the systems are tracking the 2 a.m. incidents. Chief product officer Daniel Morris frames it as a production-readiness problem more than an AI-capability one — which is a tidy way of saying the models mostly work, but the operational scaffolding to trust them at scale often doesn’t. Until those two views converge, “confidence” in enterprise AI will keep meaning very different things depending on where you sit.
