The Engineering Skills AI Agents Cannot Build for Junior Developers

Summary

Criteo engineer Elise Baturone argues that AI agents are changing how junior software engineers work, but they are not eliminating the need for engineers. Her concern is that relying on agents for both implementation and learning could leave developers permanently dependent on reviewing other systems’ output rather than becoming seniors themselves. She defines seniority through concrete behaviors: defending shipped code, explaining design choices, recognizing when an agent is going astray, deciding what to delegate, and teaching colleagues. To develop those abilities, she experimented with a `skill.md` file that instructed an agent to ask understanding checks, encourage code reuse, and stop before making unconfirmed changes. The file helped, but she found that the important gains came from her own habits: asking “why,” studying senior reviewers’ recurring comments, and comparing agent-generated code with established patterns. She also recommends occasional tool-free coding to preserve autonomy and strengthen architectural and algorithmic understanding, while helping engineers judge when manual changes are more efficient than spending tokens on delegation. Her approach includes writing one’s own questions to mentors instead of copying AI-generated messages or answers. She reports that repeated practice made her open the skill file less often and helped her earn a promotion nomination earlier than planned. The article concludes that curiosity, emulation, autonomy and collaboration require active participation by the engineer, plus time and mentorship from human seniors.