Maddy Zhang, a senior software engineer and former Google engineer, discusses how she thinks AI changes engineering work and career preparation.
Four skills to practice
- Verification: Review an agent’s diff before running tests; predict edge cases, races and network failures, then compare your predictions with the results.
- Judgment: Keep a decision log recording alternatives, constraints and why you chose a particular design.
- Managing agents: Try a small pipeline in which one agent implements a spec, another reviews the diff, and tests gate the merge.
- Communication: Write a pull-request description before implementation and keep decisions accessible to both people and agents.
Fundamentals and career evidence
- Database indexes, networking and shared state still help reviewers recognize failures that generated code can conceal.
- For a portfolio, Zhang recommends showing a spec, tests and decision log rather than relying on a long list of repositories.
- She cites company and survey data to argue that review and coordination consume more attention as code generation accelerates. Those samples describe particular organizations or respondents, not a forecast for every engineering job.
- Her title suggests learning something instead of coding; her advice actually retains programming fundamentals alongside these four skills.
“The burden is now on the person who needs to check that code, decide what’s worth building in the first place, steer the agents doing the work, and get other humans aligned around it.”