June 7, 2026

Trust issues, now fully automated

My automated doubt development process

Coder fights AI chaos by making robots doubt each other — and the comments are split

TLDR: A developer says AI can be useful for writing software only if you surround it with other AIs that constantly challenge its work. Commenters were torn between calling it the future, warning it creates ridiculous overkill, and joking that we’ve invented machines to do our doubting for us.

One developer’s big idea is deliciously paranoid: if artificial intelligence is going to help write software, then other AIs should immediately turn around and question everything. The article lays out a process where different bot “helpers” poke holes in plans, hunt for missing details, and basically act like a panel of tiny nitpicky auditors before any real work begins. It’s less “trust the machine” and more “make the machine defend itself six times before lunch.” For readers outside the tech bubble, the point is simple: the author got burned by letting AI do too much too fast, so now they’ve built a system of automated second-guessing to catch bad ideas early.

And oh, the commenters had thoughts. One camp was nodding along hard, calling this kind of adversarial collaboration the only realistic way to get decent results from AI tools. Another crowd immediately went full skeptic: why are we obsessing over how to build the feature, they asked, instead of asking whether the feature should exist in the first place? That critique landed like a reality check. Then came the funniest warning shot: one commenter said this all starts sounding like AI overengineering weird edge cases nobody cares about, unless you add a final “skeptical pass” to cut the nonsense.

The comedy gold, though, came from the comparison to Douglas Adams’ Electric Monk: a machine that believes things for you so you don’t have to. That joke basically became the thread’s whole vibe — modern coding, but make it existential.

Key Points

  • The article presents an AI-assisted development workflow designed to rebuild trust by systematically applying automated critique to specs and other artifacts.
  • The workflow relies on specialized subagents in Claude Code that review a specification from different perspectives before implementation begins.
  • Initial design review uses agents such as Pre-Implementation Architect, Documentation Validator, and Assumption Excavator to identify scope issues, documentation gaps, and hidden assumptions.
  • For broader projects, additional agents such as Gap Analyzer, Implied Completeness Detector, and Ambiguity Mapper are used to find omitted behaviors and undefined edge cases.
  • After subagent findings are incorporated into the specification, the author manually reviews the updated spec and has Claude generate a separate checklist to guide implementation.

Hottest takes

"all it takes is a 'skeptical pass'" — m12k
"much more emphasis on implementing the feature than on questioning if it's required" — docheinestages
"Electric Monks believed things for you" — watersb
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.