The agent, not the app, is becoming the primary interface for work. On June 2 at Build 2026, Microsoft made the shift explicit, launching Scout, an always-on agent that acts on your behalf without being asked. Every major platform is telling you the same thing: agentize now. The same week, the opposite signal landed. In Sinch’s The AI Production Paradox, a survey of 2,527 senior decision makers across ten countries, 74% of enterprises have already rolled back or shut down a live AI agent after deployment, anything from disabling a workflow to retiring the agent for good.
Why It Matters
The number that should stop you is not 74%. It is 81%. That is the rollback rate among organizations with the most mature governance. The best-prepared teams reversed most, because real oversight sees the misalignment casual adopters never detect. Maturity does not prevent the failure. It reveals it. This is customer-communications agents specifically, a leading indicator, not a verdict on every use case. Not every rollback is a failed agent. Some are a line set wrong, caught late. Meanwhile 98% are increasing their AI communications investment even as live deployments come back out. The gap between buying AI and running it in production is where the cost hides.
The Decision
The question this week is not whether to adopt agents. Every platform just settled that for you. It is when. A reasonable CEO can move first to capture the advantage, or sequence deliberately: pilot quietly while earlier movers map the failure modes in public. Their rollbacks become your design notes. But waiting is not protection on its own. The supervised pilot is. Speed and timing are not the same bet.
What To Do This Week
- Name the one agent your team most wants in production this quarter. Ask the owner what breaks on Friday if it misfires: a customer relationship, a compliance line, a day of revenue.
- Convert that one from a launch into a 30-day supervised pilot with a kill switch.
- Ask your COO for the rollback plan before go-live. If reversing it would be messy, you are not ready to deploy it.
What Not To Do
The risk this week is misreading the signal. Don’t read “the disciplined are reversing” as “wait and do nothing.” A supervised pilot is motion. Deploying to production fast is the gamble. Don’t treat a vendor making agents the default as your reason to put them live. The default in the product is not the default for your operation. Don’t assume governance makes you the exception. Maturity buys faster detection, not immunity.
Signal Boost
Sinch, The AI Production Paradox. The clearest dataset we have on where agent deployments break. Read it before your next agent goes from pilot to production, not after.