The Org Chart Will Eat Your AI Strategy

Someone from the C-suite just came back from a conference. Maybe they read something that scared them. The message lands clean: we need to move faster on AI. We need to be an AI-first company. We’re falling behind.

And then everyone walks back to the same teams, the same approval chains, the same quarterly planning cycles — and tries to figure out how to bolt AI onto a system that was designed, at a fundamental level, to resist change.

This is the problem. And it’s more serious than most leaders want to admit.

Conway’s Law is almost fifty years old and still not taken seriously enough. The observation that systems reflect the communication structures of the organizations that built them is cited at conferences and then promptly ignored when it comes time actually to reorganize anything. Because reorganizing is painful, political, and doesn’t show up cleanly on a roadmap.

But here’s what’s changed: the cost of this mistake has skyrocketed.

When software was expensive and slow to build, organizational dysfunction could hide inside the friction of production. You couldn’t easily tell whether your bloated integration layer was a consequence of bad architecture or bad org design. It took years for the debt to surface. By then, people had moved teams, changed jobs, rewritten history.

AI collapses that timeline. Now you can scaffold something in an afternoon that would have taken a quarter. The lag between decision and artifact is nearly gone. Which means the shape of your decisions — who makes them, how they travel, what friction they encounter — shows up in the product almost immediately.

The org chart compiles faster now.
There’s nowhere to hide.

The teams closest to the models, the people who understand what’s actually possible, what’s brittle, what requires careful judgment or human review, are often structurally isolated from the teams making product decisions.

You get AI capabilities bolted on rather than built in. Features that look impressive in demos and feel incoherent in use. Missed opportunities that nobody flagged because the person who knew enough wasn’t in the room where sequencing decisions got made.

This isn’t a talent problem.
The people are there.
The insight exists.
It just doesn’t travel.

Most organizations respond by building a centralized AI center of excellence. The COE publishes internal APIs, runs workshops, and builds capabilities. Product teams attend, get excited, then return to their backlogs. And the AI sits unused or gets integrated in shallow, low-leverage ways because the team doesn’t have the autonomy or context to use it well.

The judgment required to use these tools effectively can’t be centralized. It has to live close to the problem.

Generative AI makes this worse in a specific way. It has a seductive quality. It appears to reduce complexity while actually increasing it downstream. A team that’s overwhelmed will reach for AI to paper over the overwhelm. They’ll use it to go faster in the wrong direction. The structure has to support the kind of deliberate thinking that decides what to build before AI acceleration helps you build it better.

The opportunity is real, and it’s not evenly distributed.

Organizations that build AI into how their teams are actually structured, not just into their tools or their pitch decks, will pull ahead. Every iteration teaches them something their competitors aren’t positioned to learn. Every shipped product sharpens the next decision. The structure enables the learning, and the learning becomes the moat.

The rest will spend the next five years running pilots that don’t scale and wondering why. They’ll have the same tools, the same models, access to the same APIs. But the right people will be in the wrong rooms. Software will keep encoding the wrong assumptions about risk, value, and user behavior — not because of bad intent, but because of who wasn’t in the conversation when direction was set.

Designers have fought this battle for decades — pushing to be upstream of decisions, to shape problems before solutions get locked in. ML engineers and data scientists are fighting the same battle now. If the people who understand what these models actually do, rather than what they appear to do in a demo, are consistently downstream of the decisions that matter, the decisions will be consistently worse.

The tools are not the bottleneck. They haven’t been for a while.

The question is whether your structure allows your people to use those tools with judgment. With enough context, with the autonomy to make real calls, with the cross-functional support to make them well. Most organizations don’t. The permission layers are too heavy. The COE is too far from the work. The right people are in the wrong rooms.

While most leaders are focused on which models to license and how to measure AI ROI on a spreadsheet, that structural gap is quietly becoming the whole game. The organizations closing it aren’t doing anything exotic. They’re just taking something most leadership teams treat as an administrative detail seriously.

Your org chart is upstream of your codebase. In an AI-accelerated world, it’s upstream of nearly everything.

Agustin Sanchez is the author of The Gap — a publication about what happens when design meets organizational power, and what it costs when they don’t understand each other.