A simple A/B test to expose coding agent limits: no manual edits allowed
Liu_eroteme · x · 2026-09-28
Continuing the debate on whether senior devs' knowledge is depreciating, Liueroteme proposes a concrete experiment: pair with a junior developer well below your level, explain what you want, and have them vibe-code the PR with zero manual edits—you do the same. Both sides may review and propose changes but never edit code by hand. Comparing the two PRs, he says, reveals the gap immediately. The underlying argument: new devs start at a higher abstraction and ship fast with agents, but only deep stack understanding lets you push agents to and beyond their limits.
More from coding & agent
- AI code-sandbox provider Daytona officially moves to new domain — glcst · 2026-09-28
- Base44 launches Base Code: AI branches and PRs on your existing codebase — thetripathi58 · 2026-09-28
- YC F26's WonderSearch searches millions of documents without pre-embedding the corpus — KlausCodes · 2026-09-28
- Meta's Muse agent reportedly gives users 100M free tokens weekly plus a VM — FuSheng_0306 · 2026-09-28
- fframes: dev opens up 5-year-old video rendering framework that renders in 36 seconds — alexcovo_eth · 2026-09-28
- ThreeJS demo site threeui.com wows with stunning visual effects — vista8 · 2026-09-28