Lately, I've been working through a sequence of ideas about what AI transformation actually requires.

First, I wrote about the queryable organization: the idea that AI becomes more valuable when a company can read, retrieve, and synthesize what it already knows.

Then I wrote about the closed-loop organization: the idea that AI only starts to compound when what the organization learns makes its way back into how the business operates.

What I want to point out now is that these two ideas are connected. A queryable organization can access its knowledge. A closed-loop organization can turn that knowledge into compounding intelligence. But I think there's a third question you also have to ask: Is the organization loopable?

That is, has the company designed its work so people and AI can improve the system every time the work runs?

While I can certainly understand why, I think the dominant conversation about AI is still too focused on tools. People want to know which model to use, which platform to buy, which chatbot to approve, which department should pilot which application. Those questions are not irrelevant, but they are downstream of a more important one: can your organization actually learn from the work AI is helping it do?

If the answer is no, AI will still produce value, but the value will remain episodic. The work might be faster, but the organization will not necessarily become smarter because of it.

That’s the difference between AI as a productivity tool and AI as a learning layer. And one is definitely better than the other.

The Two Kinds of Loops

Thanks primarily to OpenClaw creator Peter Steinberger, the term “loops” is having a moment.

In engineering and agentic AI, a loop is a cycle in which a system works toward a goal, evaluates what happened, adjusts its next step by re-prompting itself, and continues. The agent works towards its goal through a sequence of acting, observing, reviewing, updating, and then trying again. The agent “loops” until it’s accomplished the goal.

But an agent can loop through a task all day long, accomplish the goal, and leave the organization no smarter if the context, work, or insight disappears when the task is complete.

For AI to compound its effect, the organization around it has to be designed for learning.

From Queryable To Loopable

A queryable organization is readable. It has made its knowledge accessible enough for people and AI to find and use. Customer history, project notes, decisions, proposals, feedback, operating practices, and institutional memory are organized into a usable knowledge layer.

A closed-loop organization is current. It does not merely collect information. When a meeting happens, a project ends, a customer raises an objection, or a field team solves a problem, the learning makes its way back into that knowledge layer.

A loopable organization is improvable. It designs work so that current knowledge improves the next cycle. The next sales call is better because the last one was captured. The next project starts smarter because the last project taught the system something. The next estimate reflects what the company has learned before.

That last step is where the real transformation begins, but most organizations were designed to simply complete work, not to learn from it. The sale is closed. The project is delivered. The service call is resolved. The estimate is submitted. The customer concern is handled. Then everyone moves to the next thing.

Inside that motion, the company creates enormous intelligence. A field team learns why a job succeeded. A salesperson learns what the customer was really buying. An operations leader spots a pattern in delays. A project manager discovers which sequence saved time.

If there is nowhere for the learning to go, AI remains a productivity tool. When the learning re-enters the system, AI becomes a much more powerful learning layer.

The companies that benefit most from AI will be the ones whose work is structured enough to teach the system what to do next.