Legible before intelligent

AI does not transform a company. It multiplies one. And almost nobody does the boring work that decides which direction it multiplies.

I have had the same conversation so many times this year that I can see it coming in the first few minutes.

A founder or a leadership team pulls me in. They are excited. They have rolled AI out across the organization. Everyone has access, the budget is approved, the announcement has gone out. They want to talk about what comes next.

Before anything else, I ask them one question. How does your team actually make decisions today?

And the room goes quiet.

Not because they have no answer. Because the answer is uncomfortable once they say it out loud. Decisions still happen the way they always have. In one person's head. In a Slack thread nobody can find two weeks later. In a meeting with no notes. In a process that works only because one specific person is still there to hold it together.

That is not an AI company. That is a company with expensive subscriptions.

Here is the thing I have come to believe after doing this with enterprises and startups both. An AI company is not defined by the tools it uses, because every company has the same tools now. It is defined by how legible it is. How clearly its processes are written down. How well its knowledge is structured instead of trapped in people's heads. How plainly it can say why a given decision gets made the way it does.

Legibility is the input. It is the thing AI can actually work with. Intelligence is the output. And you do not get the second without the first, no matter how much you spend.

The cleanest way I can put it is this. AI is a multiplier, not an engine. It does not create order where there was none. It takes whatever is already there and scales it. Multiply a clean, legible process and you get something genuinely powerful. Multiply a broken one and you get a faster broken one.

I watch this happen in slow motion all the time.

A company's onboarding is a mess. New hires spend two weeks confused, chasing answers across stale docs and dead Slack threads. So they bolt an AI assistant onto it. Now new hires get instant, confident answers about a process that is still fundamentally broken. The confusion just arrives faster.

A sales team has no real process for chasing leads, so deals slip through the cracks. They add an AI tool to send the follow ups automatically. Now leads get contacted consistently inside a process that still does not work. The activity graph looks healthier. The results do not move.

The shape is always the same. Take something broken, add AI on top, call it transformation. What you actually built is the same broken thing with better tooling and a higher monthly bill.

The companies quietly getting real value from AI did something deeply unglamorous first. Before they touched a single tool, they made themselves legible. They wrote down how decisions actually get made. They pulled the knowledge out of senior people's heads and put it somewhere a newcomer, or a machine, could read it. They made their operations readable to humans before they ever asked a model to read them.

The ones still waiting for AI to deliver on the promise skipped that part. They bought the intelligence and never built the legibility for it to stand on.

And this is not only what I have watched up close. It is now what the data shows. MIT's 2025 study of enterprise AI found that around 95 percent of company pilots delivered no measurable impact on the bottom line, and the researchers were blunt that the cause was not the quality of the models. It was the gap between the tools and the way organizations actually work, what they called a learning gap. Gartner expects more than 40 percent of agentic AI projects to be scrapped by 2027 for the same unglamorous reasons: unclear value, rising cost, and tools bolted onto systems that were never ready for them. None of that is a model problem. It is a legibility problem wearing a technology costume.

I understand why. The legibility work is boring. It does not make a good all hands announcement. Nobody posts on LinkedIn about the week they spent writing down how their team makes decisions. It feels like overhead while a competitor is shipping flashy AI features. But it is the entire reason their AI investment will deliver and yours might not.

So if AI is not delivering for you, I would gently suggest the tool is not the problem. The substrate is. The model is doing exactly what it does, faithfully scaling whatever you fed it, and what you fed it was chaos.

Before you buy the next tool or run the next pilot, try the unglamorous thing. Pick one important process and make it legible. Write down how it really works, not how the org chart says it works. Structure the knowledge that currently lives in one person's memory. Make it readable. Then point the same tools you already have at it, and watch how differently they behave.

You cannot make a company intelligent. You can only make it legible, and let intelligence be the thing that happens next.

A question to sit with, and I would like to hear your answer. How legible is your company to someone who joined last week? Because that, far more than your tooling budget, is what decides whether AI does anything real for you.

Reply and tell me. I read every one.

Sources: MIT NANDA, "The GenAI Divide: State of AI in Business 2025" (July 2025), which found that roughly 95% of enterprise generative AI pilots produced no measurable impact on profit and loss, attributing the gap to integration and workflow rather than model quality. Gartner (June 2025) forecast that more than 40% of agentic AI projects will be canceled by the end of 2027, citing unclear business value, rising costs, and weak risk controls.