Zenn: reply to '20 Claude Code Max vs your salary' by conceding the math

This is written for mid-career software engineers who run their development through AI agents and are unsure how to bring that experience into an interview. Its premise is that the question is a stress test rather than a budgeting exercise; the described interviewer is watching for calm, a plain-language account of what the candidate contributes, and a sense of tool spending as an investment. The arithmetic is a trap in only one respect — trying to win on the details of the calculation gets no credit — and an opening in another: accepting the number is what frees the candidate to talk about results.

The candidate's answer is built around three layers, in order: deciding what to build, building a system that makes AI follow rules on its own, and taking responsibility for judgments AI cannot make. The second layer involves specifics like committing rules into automated verification, Hooks that block a commit or push when checks fail, reusable workflows shared across repositories, and past decisions to rip out a tool after measuring that it did not help. The third is reserved for areas like payments, authentication, and data migration, along with production incidents and expanding practices to a team.

One question at the end is what moves the interview forward, because it changes the question from a test into a way of choosing what to discuss next. From there it branches: if AI adoption is not yet organized, lead with building the environment and a one-month rollout plan scoped to a single team and repository; if it is already running, lead with responsibility and spreading practices, since review and judgment become the bottleneck as AI speeds up writing. It also recommends being honest about whether experience came from personal projects, and adjusting length by who is across the table.