Tech Lead Sergey Proshchaev notes that while AI agents help junior developers close tasks quickly, they often hinder the development of deep system understanding. Metrics like cycle time may appear positive, but juniors frequently struggle when real-world incidents occur because they lack the ability to diagnose complex system failures independently.
Data from the LeadDev Engineering Leadership Report 2026 shows that 84% of leaders believe AI tools negatively impact junior growth, yet only 4% have adjusted their onboarding processes. A January 2026 study by Anthropic involving 52 developers found that while AI assistants slightly increased speed, they significantly reduced performance in debugging tasks, especially for those overly dependent on AI generation.
To address these gaps, experts like Garima Agarwal suggest shifting focus from delivery metrics to skill-based assessments. Onboarding should prioritize manual system exploration, such as tracing requests from API to database, before allowing AI usage. Mentors should conduct regular reviews of past incidents to teach diagnostic skills rather than just task completion.
A structured approach involves categorizing tasks into 'delivery' and 'learning' types. For learning tasks, juniors are encouraged to attempt solutions manually, document their hypotheses, and then compare their approach with AI suggestions. This method ensures that developers build a mental model of the system while still learning to use modern tools effectively.
