A former paralegal turned engineering manager uses his courtroom‑assembly‑line past to warn against a new management fad called tokenmaxxing. The practice rewards developers for the sheer number of LLM tokens they consume, turning AI adoption into a leaderboard game. He argues the metric is as useless as a stopwatch‑watching executive once was.

To counter the hype, the manager drafted an internal AI policy that eschews mandates and focuses on accountability. Engineers must understand any code the model generates, be able to work without the tool, and prioritize customer outcomes over token counts. Senior‑level staff are encouraged to experiment, but performance reviews will not track usage.

The policy also stresses that AI‑generated code remains the developer’s responsibility; reviewers should not be burdened with unchecked submissions. When tooling improves, teams can adopt it, but the underlying codebase must stay maintainable for the next decade. In practice, the document gives engineers a clear framework rather than a scoreboard.

By refusing to tie compensation to token consumption, the team avoids the perverse incentives that plagued the earlier stopwatch experiment. Senior engineers can now choose tools that genuinely speed delivery, while junior staff receive guidance on safe AI use. The result is a pragmatic stance that treats AI as an aid, not a mandate.