Nearly every company I walk into has an AI pilot running somewhere. Very few have an AI operating model. That gap, between a pilot and the way the company actually runs, is where most of the value, and most of the failure, lives.
McKinsey's State of AI survey puts the shape of the problem in numbers: 88% of organizations now use AI in at least one function, yet roughly two-thirds remain stuck in the experimenting or pilot stage. Adoption is not scaling, and I have argued before that the foundations come before any magic.
Pilots are a starting point, not a strategy
Copilots are a mere starting point: they bring visibility to AI through chat as a first touchpoint, but they do not, on their own, lead to meaningful change. The real work begins when you rewire the operating model around them, connect the data, and turn scattered pilots into a coordinated capability the whole organization can build on.
Build the digital twin sideways
One of the most important moves is to build a digital representation of your business, a digital twin, sideways, rather than trying to re-integrate 80 or 90 fragmented systems. As I explained in my note on the new operating model, this horizontal abstraction layer bypasses the decade-long struggle of integrating fragmented legacy systems and provides the unified context AI needs to operate.
Agentic loops become the operating core
The new model is built on agentic loops: models that use computer tools natively, navigate systems, and execute work end-to-end. This is what I called in my note on narrow to generative AI the shift from chatbots to agentic loops that control all digital tasks on a computer. It is what drops the cost of cognitive labor toward zero for those who adopt these architectures.
An operating model is not a project
An AI-ready operating model is not a project with a deadline. It is the combination of foundations, a sideways digital twin, agentic loops, and leadership that treats AI as the way the business runs and not something bolted on to it. If you want to see the failure mode up close, I documented it in why most AI transformations fail.