Has AI Conquered Coding? (It’s Not So Simple…) | AI Reality Check

Deep Questions with Cal Newport

Has AI Conquered Coding? (It’s Not So Simple…) | AI Reality Check

Deep Questions with Cal NewportMay 21, 2026

Why It Matters

Understanding the limits of AI coding tools is crucial for tech teams aiming to maintain code quality and developer growth, especially as AI becomes ubiquitous in software pipelines. This episode offers a timely caution against unchecked automation, helping listeners navigate the trade‑offs between speed and long‑term skill sustainability.

Key Takeaways

  • AI coding agents speed up tasks for skilled developers
  • Overreliance erodes critical thinking and debugging skills
  • Junior developers risk missing foundational coding experience
  • Balanced approach: use LLMs for specs, write core code manually
  • New productivity metrics may increase burnout and low-quality code

Pulse Analysis

Recent buzz around AI‑driven software development centers on an essay by programmer Lars Faye, who warns that the “agentic coding” hype oversimplifies reality. The essay describes a workflow where developers define requirements, then repeatedly pull a “slot‑machine lever” of AI agents to generate code, creating a widening gap between the orchestrator and the actual codebase. Proponents claim AI can turn English prompts into production‑ready software, promising a future where traditional coding disappears. The episode unpacks this vision, questioning whether AI can truly replace the deep architectural thinking that seasoned engineers provide.

Faye’s core argument highlights a growing skill‑atrophy problem. He cites Reddit posts where seasoned developers admit their ability to spot bugs and understand syntax is deteriorating because they rely on AI to write most of the code. Junior programmers face an even sharper “junior‑year wall”: they skip the essential struggle of writing code from scratch, leaving them unable to debug or reason about generated solutions. A veteran with three decades of experience confirms these trends, noting faster output for experts but also chronic context‑switching, mental fatigue, and the emergence of new, misleading productivity metrics such as token counts that mirror the outdated lines‑of‑code KPI.

The proposed remedy is not to abandon AI tools but to demote their role. Faye recommends using large language models primarily for specification, planning, and pseudocode, while writing 20‑to‑100 % of the actual implementation oneself, especially for critical components. This hybrid approach preserves core programming competence, enables meaningful code reviews, and prevents burnout caused by endless agent‑driven iteration. As AI coding assistants become ubiquitous, organizations that balance automation with continuous skill development will likely maintain higher code quality and healthier teams, while those that over‑automate risk a new wave of productivity‑driven decay.

Episode Description

Cal Newport takes a critical look at recent AI News.

 

Video from today’s episode: youtube.com/calnewportmedia

 

(0:00) Has AI conquered coding?

(3:21) Lars Faye quote

(5:25) Skipping the struggle step

(6:42) Point #1

(7:08) Point #2

(7:28) Point #3

(7:39) Point #4

(8:35) Solution

 

Links:

Sign up for Cal’s newsletter at www.calnewport.com/ideas

Buy Cal’s latest book, “Slow Productivity” at www.calnewport.com/slow

https://larsfaye.com/articles/agentic-coding-is-a-trap

https://www.infoworld.com/article/4143101/pity-the-developers-who-resist-agentic-coding.html

https://www.youtube.com/watch?v=OQSNhk5ICTI

Thanks to Jesse Miller for production and mastering and Nate Mechler for research and newsletter.

Learn more about your ad choices. Visit podcastchoices.com/adchoices

Show Notes

Comments

Want to join the conversation?

Loading comments...