Why the Suits Aren't Seeing ROI on AI

I’ve worked at an enterprise for a while now, and as you’d expect, everyone and their mom is trying to figure out how to use AI for efficiency gains and profit. Which is fair. Of course they are.

After reading McKinsey’s State of AI report and the Anthropic Economic Index, one thing got clear to me: a lot of orgs aren’t seeing ROI on their AI projects.

I think part of the problem is the organizational politics that come with any big company. My reasoning is simple. Everyone is out for their own interest, not necessarily the company’s. That’s a human trait, so it makes sense.

Here’s the thing about AI specifically. It’s a fast, new, constantly moving field. Anyone without a real passion for it is, in some sense, not qualified for the job. And yet plenty of leaders who’ve never run a single instance of Claude Code or Codex are the ones signing off on complex AI architectures.

You might see where I’m going.

To seek out extraordinary talent and have good judgment about what to build and what to skip, you have to be on the front lines yourself. When you’re not, you get a recursive loop: a manager tells a boss who’s also clueless that they know how to “do” AI, and the boss has no way to check the claim. He has no baseline.

So the charlatans become the bottleneck. They never get fired, because the boss and the boss’s boss can’t tell good from bad. They don’t have a reference point.

Meanwhile the ICs and leads have a much better read on what’s good, what’s bad, and the second and third order effects of an architecture decision. The problem is the layer between them and the top. That middle layer is where the great filter happens.

So what’s the fix?

I like what Jack Dorsey is doing at Block. After cutting about 4,000 jobs (roughly 40% of the workforce) in February, he published an essay with Sequoia’s Roelof Botha called “From Hierarchy to Intelligence” laying out the model: no permanent middle management, just ICs, people who own specific problems, and “player-coaches” who build product and develop people at the same time. His argument is basically mine: hierarchy exists to route information through orgs too big for one person to oversee, and AI now does that routing. Block is on the leading edge relative to legacy enterprises, but I think this is where we’re all headed. We don’t need that many people anymore.