Most AI Failures Are Not AI Failures A regional bank in the northeastern US came to us with a familiar ambition. Move from a hierarchical structure to a project based operating model, faster decisions, less layered approval, technology and operations working as one team instead of two. The technology side was never the hard part. The hard part was that nobody had touched decision rights. People kept reporting the way they always had, escalating the way they always had, getting evaluated the way they always had. The bank wanted agility without redesigning who owned what, and that gap is where the actual work began. We ended up redesigning the operating model for the technology and operations group, building a new talent platform and reward framework around it, because the structure had to change before any process inside it could. This is the pattern we keep seeing, and it rarely gets named correctly. Pilots succeed on a narrow, well defined task with a small group of engaged user...

Claude Managed Agents: The Infrastructure Moment Agentic AI Has Been Waiting For There is a moment in every technology cycle when something shifts from being a proof of concept to becoming infrastructure. We saw it with cloud computing when Amazon Web Services moved from developer curiosity to enterprise backbone. We saw it with mobile when the iPhone stopped being a device and became the primary computing surface for billions. We are watching that moment happen again with agentic AI. The public beta launch of Claude Managed Agents by Anthropic is one of the clearest signals yet that this shift is underway. Not because of what was announced, but because of what it removes, and what it exposes. For those tracking the space closely, the announcement itself is not surprising. What matters is what it represents structurally, and what it demands from enterprise teams that have so far treated agentic AI as something to evaluate rather than something to govern. Claude Managed Agents allows ...

There is a moment in every technology deployment cycle when scale transforms a manageable problem into a structural one. A 5% error rate on a hundred decisions is five mistakes, each recoverable, each visible, each correctable before it compounds. A 10% error rate on 5 trillion decisions per year is something categorically different. It is tens of millions of wrong answers delivered every single hour, each one formatted with exactly the same visual authority as a correct answer, surfaced to users who have no mechanism to tell the difference. A The New York Times investigation into Google’s AI Overviews put that number in the open. The 90% accuracy figure sounds reassuring until we do the arithmetic at scale and then ask the harder question the accuracy frame keeps avoiding. The harder question is not how often the system is right. It is whether the system has any accountability architecture at all for the moments when it is wrong, and whether that architecture is visible to the perso...