March 23, 2026
AI ships code, humans ship takes
How I'm Productive with Claude Code
Genius hack or manager cosplay? Devs spar over AI “interns”
TLDR: An engineer automated coding busywork with Claude Code and ran multiple tasks at once while acting as a “manager.” Commenters split between fans calling it a superpower and skeptics warning about credit, bad metrics, and risky oversight—asking who’s responsible when AI “teammates” get it wrong.
One engineer says he turbo‑charged his job with Claude Code—an AI “agent” that writes code, opens pull requests, previews the app, and even checks its own work—while he plays manager and keeps five features cooking in parallel. He cut wait times to near zero, automated the boring bits, and claims the real win is a quiet mind and nonstop flow.
The crowd? Split and loud. Fans like jmathai cheer, boasting the same setup and confessing “the workflow is so good, I’m the bottleneck.” Others rolled their eyes at the boss vibes. CrzyLngPwd roasted the “good manager” line with a zinger about managers “claiming credit” and asked who gets punished when the “junior” is an AI that misunderstands instructions. markbao admired the buildout but warned the fun could fade when the job becomes endless review duty. And serf slapped a big red sticker on the victory chart: commit graphs are a terrible productivity metric—like counting lines of code and calling it quality.
Jokes flew about “AI interns that never sleep,” “commit confetti,” and “manager cosplay.” Skeptics like MeetingsBrowser cautioned the whole flow assumes AI can finish tasks with minimal oversight. Love it or hate it, everyone agrees: this isn’t just about code—it’s about credit, risk, and what actually counts as getting things done.
Key Points
- •A Claude Code skill (/git-pr) was created to automate staging, commit messages, PR descriptions, and GitHub PR creation.
- •Switching the build to SWC reduced server restart times to under one second, preserving developer flow.
- •Claude Code’s preview feature enabled agents to verify UI changes themselves, reducing bottlenecks.
- •A system assigning unique port ranges per Git worktree eliminated port collisions and enabled parallel previews.
- •The author scaled to running multiple agents across several worktrees, focusing on planning and final code review.