A engineering manager shares the AI policy he wrote for his skeptical team, rejecting token-count mandates in favor of code ownership and people-first values.
Key Takeaways
“Tokenmaxxing” treats token consumption as a KPI; engineers trivially game it with loops, making it a vanity metric divorced from customer value.
Policy has no AI mandate: engineers are not reviewed on tool usage, but are expected to stay aware of a rapidly evolving space.
Any AI-generated code is the author’s code; engineers must understand it, fit it to existing patterns, and not shift review burden onto teammates.
Junior engineers should use AI tools judiciously because learning happens through struggle and reps; outsourcing code writing to an LLM stunts career growth.
The codebase has a decade of history and product-market fit; AI maximalism that accrues tech debt faster than models improve is an explicit bad bet here.
