Rivedere il codice automatico non è la soluzione neppure per gli junior

Summary of the video

The speaker discusses a follow-up blog post and the debate it sparked about using AI to write code. Their main argument is that “re-checking/reviewing” code generated by large language models (LLMs) is not a real solution—especially for junior developers—and that obsessing over AI-generated code can limit how people actually benefit from it.

They compare this to a broader work dynamic: if someone doesn’t trust others’ output and constantly monitors it, work becomes inefficient and people stop progressing. Similarly, developers who don’t trust outputs (or focus too heavily on inspecting generated code) may not delegate effectively to AI as a leverage point.

Key points

• AI-generated code can be good or bad; reviewing it doesn’t automatically guarantee quality or correctness.
• Junior developers shouldn’t learn primarily by looking at AI output. Learning programming usually requires actively writing code by hand and understanding underlying concepts.
• AI can still help juniors unblock themselves when they’re stuck—especially if they don’t understand how to interpret books/papers or don’t know what mistake they made.
• A better use of AI is mentor-like assistance, not a replacement for the learning process.
• For experienced programmers, reading AI-written code can be valuable only in special cases; otherwise, reviewing it may be less productive than writing key parts yourself to truly learn.
• Even though AI changes workflows, it doesn’t remove the need to understand what’s happening. Over time, the best programmers will likely remain those who understand the system deeply.
• The speaker predicts a potential generational split: some newer developers may write software with less understanding, potentially leading to weaker outcomes than developers who know what’s happening “under the hood.”
• A hopeful twist: AI may make learning programming more broadly worthwhile, even for people who don’t want to become professional software developers. Many areas (like research and mathematics) may require only a basic “smattering” of programming to build meaningful software, rather than full mastery from the start.

Speakers

• Primary speaker (unnamed): the person narrating the discussion throughout.
• Leonardo di E… (name truncated in subtitles): referenced as someone who picked up the blog post and explained his reasons clearly (no direct quote from him appears in the subtitles).