Summary
Highlights
The speaker, a Meta engineer, reveals he hasn't written a single line of code in six months, a trend observed in the industry with figures like Andrej Karpathy and Boris Cherny. He questions if coding is still important in 2026, highlighting the increasing role of AI in writing production code.
Instead of writing code, the speaker now spends most of his time inspecting and evaluating the AI's logic, output, and generated tests. He argues that 'judgment and taste' – knowing what good code looks like and steering AI – is the true value of an engineer today. He uses analogies like a chef or a music producer (Rick Rubin) who excels without directly performing the core task but possesses critical discernment.
The video delves into whether a beginner without hands-on coding experience can develop the necessary judgment. While one can develop 'taste of discernment' by studying codebases, the ability to fix issues when AI fails still requires fundamental coding skills. The speaker contends that while AI is rapidly improving, it's not yet at a level where engineers can completely forgo understanding code.
The speaker explores different viewpoints on AI's role, from Steve Yegge's extreme embrace of AI coding to DHH's cautious approach. He places his own current stance in the middle: engineers no longer need to type every line, but they must read, review, and thoroughly understand all code they ship.
He proposes a new 'skill ladder' for engineers: writing code (which AI can now largely handle), reading code (crucial for junior engineers), verifying code (thinking about edge cases and tests), steering AI (guiding AI to optimal solutions), and understanding intent/judgment (viewing code as part of a larger system or product). The focus shifts from the nuances of language syntax to higher-level understanding and critical thinking.
The speaker emphasizes building real projects with AI. He introduces a learning loop: pick a small, personally relevant project; build it in small, manageable pieces; read and understand every line of AI-generated code; use AI to explain confusing parts; purposely break the code to understand its limits; and quiz yourself on your understanding. This process, though slower, is vital for developing strong engineering fundamentals in the AI era.
The speaker concludes by mentioning a potential project: a 'student AI coding harness.' This tool would guide learners through the described loop, progressively challenging them and assisting with quizzes to deepen their understanding of AI-generated code. He invites interest from viewers for this potential project.