Every company I talk to is running the same play right now: document your process, define what “good” looks like, then point AI at it and let it repeat, and compound.
We’re working through this ourselves, so none of this is written from a safe distance.
Small problem. Most companies were never process companies to begin with.
Repeatability makes sense for manufacturing, insurance or procurement. But high-agency (in the sense of independent action, not the agency business model) knowledge work is different: variation isn’t a defect, it’s often the source of value.
High agency work runs on autonomy, flexibility and range. Two senior designers or developers can work in completely different ways and both be excellent, exemplary, rewarded. Multiply that by a whole team of unique cases, forming and then performing.
Motion Efficiency Study, c. 1914. National Museum of American History, Behring Center, Division of Work and Industry Collection
Claire Hughes Johnson comes at this from the idea that you should provide more context than you think you need, rather than trying to exert more control - some structural boundaries, but you can’t over-constrain people.
Git is a good example: almost no one tells you how to write code, but everyone agrees on commits, pull requests, and reviews (git is also the worst form of version control except for all the others).
Success gets measured on the quality of the outcome, and the culture makes room for variety in how you get there. It works. It’s worked for decades. Its made some companies very profitable.
AI rewards the exact opposite: consistent inputs, consistent ways of working, a documented definition of “good.” Code-like stuff.
Claire Hughes Johnson’s philosophy was developed for human organizations. AI changes some of the assumptions those organizations were built on.
AI only gets leverage on what’s legible.
So AI is exposing inconsistency that companies haven’t just tolerated for years. They’ve celebrated. And that’s not hubris, the looseness was the strength. Nobody was wrong to build this way.
Here’s the part that I think gets missed.
Becoming process-driven is a full organizational transformation on its own. Companies attempt it all the time, with consultants and steering committees and change-management budgets, and they still fail at alarming rates. Michael Hammer was writing about why back in 1999 : companies redesign the process but keep the functional org chart intact, so the process cuts across departments while authority, incentives, budgets, and careers stay inside silos. The process ends up being nobody’s job.
It’s genuinely hard: it means picking a few shared ways of working and deliberately letting the others go, which is exactly the kind of choosing-on-behalf-of-everyone that autonomous cultures exist to avoid.
Now companies are attempting organizational transformation and an AI transformation at the same time. Either one alone is big enough to break most organizations. Now companies are stacking them.
And AI’s speed removes the cushion that used to make looseness affordable. When a cycle took three weeks, friction got absorbed in the slack - a fuzzy handoff here, a “who owns this?” there. Compress the cycle to a day and coordination overhead becomes most of the delivery time. The flexibility starts costing.
I don’t think the answer is “become a process company.” AI dissolves structure at the execution layer even as it demands structure where things have been loose: shared intent, a defined “good” for the things that matter, clear ownership. Stay loose where autonomy still pays; get tight where looseness got expensive.
But that framing hides the uncomfortable part. Getting tight means someone decides - this format, this sequence, this tool or model, this definition of good - and everyone else gives up their version, which was probably also excellent (again where Git is a good mental model).
I don’t think most companies have answered the process question, honestly. Let alone noticed that the AI question is gated on it.