We went all in on AI early. We declared tokenmaxxing as a strategic posture. We invested heavily in training. We signed contracts with Anthropic, Google, and OpenAI. The token curve did what every adoption curve does at first: it climbed fast. Now it is flattening.
The instinctive reaction is to read saturation as "the tools gave us what they could give". They did not. The wall is not a model limit or a use case limit. It is structural, and it lives in the org chart.
Take one of our domains: twenty-one people cut along specific technologies and specific audiences. That structure was correct when the work required twenty-one people. It is probably not correct now. Five people plus the right agentic pattern can plausibly cover what twenty-one covered. The token curve flattens because nobody on a twenty-one-person team can actually use more tokens. The org structure is the bottleneck.
Once you accept that you have to refactor the org, the questions stop being about AI and start being about everything else. How small should teams get. What do those smaller teams actually own. Is it worth refactoring a ten-year-old system just because it now becomes feasible. How does any of this turn into revenue, not just lower cost.
Our token chart is no longer telling us anything about AI adoption. It is now telling us that parts of our organization are ready to be refactored. Tokens stop climbing because the org stopped fitting the work.
Thoughts? Find me on Bluesky.