Right-level your AI with the CAI framework

Is Mrs. Jensen's pug eating your Enterprise AI?

My grandfather had a shed stocked for any project - timber, tools, and he also had seemingly endless patience. When I was nine and obsessed with rundbold (Danish stickball), we built a custom bat together. It was perfect for me: long, well-balanced, and satisfying to swing. The first time I took it to the field, I hit the ball clean over the back fence.

But there was a problem - the younger kids couldn't use it at all. The handle was too thick for small hands, and half the team kept dropping it mid-swing. I took it back to the shed, and tried to fix it with the saw, but my attempt made the grip too thin - it snapped on the second test swing - disaster!

The third version was grandad's idea: an old, slightly rounded chair leg for the barrel. It gave the bat a beautiful weight and real hitting distance. We were delighted - that was until we actually played. There was no control over where the ball went. Often we'd shank the shots straight into Mrs. Jensen's garden, where her pug would sprint out, tail spinning, and gleefully chew our tennis balls to pieces. We lost two balls that afternoon - a real tragedy.

Grandad eventually handed me a planer and showed me how to flatten one side of the barrel. That one small adjustment changed everything. The bat became predictable, controllable, and usable by every kid on the team. Mrs. Jensen's pug went on a tennis-ball-free diet, but was happy to watch us from the sidelines.

Deploying AI with a faulty bat

Right now, enterprise AI looks exactly like those early bat prototypes. Companies measure AI maturity by counting software licenses or the number of prompts sent per day. This measures adoption, but it fails to measure value. They are handing out powerful tools that lack structural control.

That is why I published the Cognitive Autonomy Index (CAI) framework. It acts as the "planer" for your AI strategy, designed to measure and manage the shift from human execution to system autonomy.

CAI framework details

To stop hitting the ball out of bounds you need the right tools, and to stop feeding your AI and data to the neighbor's dog you must be able to tell when the tools are working - you must treat AI as a living, breathing system.

True ROI happens when humans stop "doing the machine's homework" entirely. First, evaluate your Current Operational Level (L_c) to see how tools are actually being used:

To achieve "Right-Leveling"-aligning a tool's capabilities with human competency and operational processes-you must calculate your Efficiency Score:

• $L_c$: Current Operational Level (1-4).
• $S$: Execution Competency of the user (1-5).
• $L_t$: Utility Ceiling (the maximum safe and technically possible level of autonomy).

A score over 1.0 indicates a Governance Risk, meaning your team is pushing the tool beyond its safe, compliant limits. A score under 1.0 indicates Stranded Capability, meaning you are under-utilizing the tool.

To fix this gap, deploy AI Champions to build "Actionable Harnesses" that hitch AI to specific business processes via APIs. This moves your team from being low-level creators of inputs to high-level governors of autonomous systems.

Demonstration of CAI scores

Below is a demonstration of the different CAI scores calculated from real-world scenarios - you can select a scenario to see the score and details of how it can be improved

Enterprise AI Autonomy

The rounded chair leg looked like a great bat at first glance, but without a controlled surface, it was a liability. Handing out unmanaged AI access is the digital equivalent - make sure your understand how your enterprise can use new AI tools and procedures to the best of their abilities. An unmanaged LLM pushing past its Utility Ceiling can easily compromise your operations. The CAI framework gives your enterprise the structure it needs to safely harness AI, ensuring that when your team steps up to the plate, they know exactly where the ball is going.

Have you calculated your team's CAI score yet? How will you iterate and use the framework to ensure your enterprise AI doesn't end up feeding your data straight to Mrs. Jensen's pug?