The narrative being recycled is that AI tooling spend spiraled out of control, executives panicked, and the industry is now seeing a responsible retrenchment. That framing is wrong twice over.
The first time it is wrong is in the past tense. The bills did not go up by accident. Leadership knew exactly what unlimited Claude Code looked like on the invoice.
I know, because that's what we did.
We paid steep sums because we were buying adoption, not productivity - a deliberate spend to overcome resistance from engineers who did not want to use the tools, fast. The cost was the point. The behavior change was the asset.
The second time it is wrong is in the present tense. The rollback is not a return to discipline. It's not about phasing out agentic coding or "going back to the old ways" of writing software. The rollback is what happened when the price Anthropic charged us crossed the price we were willing to pay for a steep adoption curve. API costs went up. Unlimited subscriptions shrank. The math broke. The strategy did not get smarter. It got priced out.
That sequence matters because it predicts the next move in the software industry. The moment a cheaper option clears the adoption threshold - open-weight models, local inference, in-house pipelines - the AI budget will reappear in a new line item, and someone in leadership will frame it as a cost-saving measure, or as discipline returning. It is not discipline. It is the same investment under a different name. Only the vendor changes.
I know, because that's what we're doing right now.
Thoughts? Find me on Bluesky.