February 28, 2026
Autopilot devs, manual outrage
747s and Coding Agents
AI is coding, humans are just passengers — the comments are chaos
TLDR: An AI lab coder says agents now build entire features, making him feel like a pilot watching autopilot and worrying about skill growth. Commenters split: some defend hands-on ownership and learning, others shrug that coding was assembly anyway — sparking a debate over productivity versus craftsmanship.
A coder compares his new AI-assisted life to flying a 747 on autopilot: once thrilling, now “no improvement.” He says modern “coding agents” (AI tools that write features end-to-end) let him ship more but learn less, turning him from engineer to overseer. Cue the community meltdown. The strongest reaction? Code ownership vs convenience. User flyinglizard flat-out refuses to hand over the wheel, sticking to a half-manual workflow to keep control of the code. The spicy prediction: ownership might stop mattering at all — and people aren’t ready for that. Others argue reviewing AI’s work is a different, harder job. skepticATX says agents make them more productive but worse at their craft. omoikane counters that reviewing human code teaches more, while thesz flexes: for every line they write, they read “hundredths” — translation: a lot. The drama escalates with jokes: borzi calls the old way “like building IKEA furniture,” implying coding was always assembly, not artistry. Memes fly about becoming “passenger princesses” of your own repo and being demoted to “code janitors.” The article’s nod to Jekyll docs becomes a punchline: why learn the manual when the bot slaps the feature in? The crowd is split: ship faster or stay sharp — pick your turbulence.
Key Points
- •The author contrasts a veteran 747 pilot’s career plateau with evolving software work that once offered constant learning.
- •Historically, the author learned by manually implementing features (e.g., via Jekyll docs, plugins) and iteratively debugging.
- •Early LLM use mainly replaced search and provided quick explanations without removing the need for understanding and planning.
- •Recent AI coding agents can implement entire features, leading the author to attempt agent “one-shot” solutions first.
- •Reliance on agents increases productivity but slows personal skill growth and complicates interventions when agent code is flawed.