Automating Myself Out of Development

One coder let AI take the wheel, and the comments instantly turned into a civil war

TLDR: A developer described letting an AI assistant handle more and more of their software work, only to discover that checking the machine’s output became the new grind. Commenters split hard between hype, skepticism, and mockery, with many arguing the real problem is that AI may speed things up while also multiplying the mess.

A developer wrote about slowly handing more and more of their coding work to Claude Code, an artificial intelligence helper, until the process started to feel like they were almost managing a tiny robot office instead of building software themselves. It began with one-on-one sessions, then spiraled into multiple windows, parallel tasks, and a growing pile of notes, plans, and checklists. The promise was simple: let the machine do the boring parts while the human keeps the fun bits. But the twist? The more got automated, the more exhausting the final checking became.

That tension is exactly where the community pounced. One camp basically yelled, "show me the receipts". noelwelsh said these tools seem fine for small, simple jobs, but once real design decisions show up, they can "quickly become a mess." Another thread of commenters mocked the whole vibe as overcooked productivity theater, with one brutally dismissing it as "Yegge tier psychosis." Ouch.

But the spiciest drama came from the middle ground. yieldcrv called out the two extremes: the all-in AI evangelists and the veteran control freaks who refuse to let go. Meanwhile, gnunicorn zeroed in on the real pain point: not writing the code, but reviewing it all after the bot is done. That became the accidental punchline of the whole story: congratulations, you automated the work… and promoted yourself to full-time hall monitor.

Key Points

  • The article documents the author’s progression from direct, synchronous Claude Code sessions to more automated and parallel AI-assisted development workflows.
  • The initial workflow combined brainstorming, implementation, review, and merging in a local Claude Code session supported by markdown-based context files, skills, MCPs, and sub-agents.
  • The author later scaled the process by opening multiple Claude Code windows, using worktrees across repositories, and sometimes running work on different projects simultaneously.
  • A superpowers plugin helped structure a workflow of brainstorming, specification, planning, implementation, review, and testing with additional sub-agents.
  • As automation increased, the author encountered context-switch fatigue and continued to rely on manual approvals, while also noting security concerns around OpenClaw-related tools before AWS Lightsail offered one-click deployment.

Hottest takes

"there’s gotta be a middle ground between totally AI pilled people... and control freak veteran developers" — yieldcrv
"running multiple agents in parallel would just increase the speed at which it can stuff up the project" — noelwelsh
"More Yegge tier psychosis" — brcmthrowaway
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.