Push back on an AI-first rule for every task

Your engineering org has gone "AI-first". Leadership tracks how much code each team writes with agents, and engineers feel they are underperforming when they write code by hand. Your team owns the ledger service that records every customer's balance. Your director asks you, as the team's staff engineer, for the team's position.

Where should your team not use an agent as the main author, and how do you say so without sounding against the tools?