50% of senior AI leaders have discovered an agent accessing data it shouldn't. They also report the highest confidence in visibility into their agent systems. Those two things can't both be true. We surveyed 260 enterprise engineers and AI leaders about what it actually looks like to run AI agents in production. The numbers don't lie: — 75% of hands-on engineers say their org deployed agents before they felt ready — 63% discovered an agent accessing data they didn't know it could reach — 36% can't roll back a failing agent within minutes — 54% expect to significantly rebuild systems they've already shipped The builders and the boardroom are looking at the same systems and seeing completely different things. That's not a communications problem. It's a risk profile. The pressure to move fast isn't going away. But speed without the operational layer to catch failures isn't a strategy — it's a countdown. Full report in comments 👇 #AIagents #agentic #DataObservability
The more interesting gap here is highlighted in each graph as the delta between bars...the gap between builders and leadership. Either the builders aren't sharing the reality on the ground with leaders or the leaders aren't listening. That's a trust gap that is not about Agents. That's the trust missing from most low performing teams. In an accelerated world with Agents, that human trust gap becomes a corporate risk gorge. Human trust across the team is the foundation for human to agent trust.
This gap between perceived control and actual behaviour is where most risk sits today. What stands out is that visibility without ownership doesn’t translate into control. If no one is accountable for what the agent can access, decide, or trigger, then issues will only be discovered after the fact. The organisations that move ahead are the ones embedding governance directly into how agents operate, not just observing them.
Teams think they can see what agents are doing, but in reality they often can’t fully control or track them. That gap is where real risk shows up.
Barr Moses Important signal. Confidence without operational visibility creates blind spots. Agent adoption is moving faster than governance maturity, and that gap is where enterprise risk compounds.
This gap shows up a lot. Dashboards say “green” while agents are quietly doing the wrong thing.
Barr Moses The gap between perceived control and actual system behavior is the real concern. Agent scale without observability and rollback discipline creates operational risk faster than value.
https://www.montecarlodata.com/blog-mc-builders-in-production/