Your AI adoption problem isn't tech debt, it's the operating model

Conway’s Law says your software mirrors your org structure. For years, people joked about seeing it in JIRA workflows and microservice boundaries. Today the same pattern shows up in the skills, MCPs, and scaffolding companies bolt onto AI to make their legacy codebases work.

The problem is not the codebase, and it never was. I’ve talked to enough of these companies to see the pattern. They cling to old structures, routines, and job descriptions. The planning rituals assume quarterly cycles that made sense when shipping was slow. The job descriptions hire for skills that describe yesterday’s constraints. The reporting lines optimize for coordination overhead that newer teams have eliminated entirely. Every part of the organization reinforces the shape it already has. These layers were built to protect the scarcest resource in the building: focused engineering time. That resource isn’t scarce anymore. The people inside aren’t the problem. The container is.

This is what makes the AI conversation so painful to watch. These companies adopt AI because they know they need to move faster. But they layer it onto the same broken patterns and assumptions that created the problem in the first place. AI follows existing code patterns and amplifies them. It doesn’t evaluate whether those patterns are good. It just reproduces them, faithfully, at speed. A clean architecture gets compounded into something powerful. A messy one gets messier, with more confidence, in less time. The companies trying hardest to catch up are accelerating in the wrong direction. They’re automating their dysfunction and calling it transformation.

That’s the trap. It’s not that incumbents can’t access AI. It’s that their organizational structure ensures AI reproduces the exact problems they were trying to solve. The tool works. The context it’s deployed into is broken.

Meanwhile, companies that were greenfield three or four months ago have already built out at a scale that used to take years. Clean architectures from day one, so AI compounds on solid foundations instead of fighting them. Small teams producing at a pace incumbents can’t match with 10 times the headcount, not because the engineers are better, but because the structure never got in the way. These teams didn’t overcome the structural trap. They never entered it.

The only remaining advantage is distribution: existing customers, established channels. That’s real, but defending it requires the same organizational adaptability that these companies have already proven they don’t have. The trap doesn’t stop at the codebase. It extends to every part of the business that needs to move.

And the structure won’t move itself.