It’s so hard to finish an idea that is not yours and is just suggested by AI

Original Article ↗ Hacker News Discussion ↗

AI assistants are increasingly woven into coding and note‑taking workflows, but many users report that letting models generate ideas, comments, or notes undermines their sense of ownership and makes systems harder to understand over time. Participants describe problems like hallucinated code comments turning into misleading “documentation,” AI-generated text polluting personal knowledge bases, and a growing temptation to think about how to prompt the model instead of deeply understanding the underlying problem. Some advocate strict separation between human and AI content or limiting AI to search, transcription, and lightweight summaries, arguing that the real value of notes and design work lies in the human thinking required to produce them.

AI-Generated Ideas and Ownership

  • Many find it hard to execute on ideas “suggested by AI” because they lack personal ownership and internal reasoning behind them.
  • Some say it’s equally hard to finish their own ideas; the problem is about execution in general, not just AI.
  • A cited study suggests when people believe AI did the creative work, they feel less meaning and put in less effort.
  • Several argue AI can’t truly “have” ideas; it samples plausible options without real preferences, so its “choices” often feel arbitrary.

AI in Codebases: Comments, Specs, and Error Propagation

  • Strong criticism of LLM-written code comments: they hallucinate design intent, get treated as canonical later, and poison future work.
  • Some teams now aggressively delete or block AI comments, or add harness rules to limit and validate them.
  • Others defend comments for documenting “why” and external constraints, but agree LLMs rarely produce that kind of comment reliably.
  • There’s concern that AI-generated documentation and comments can look authoritative while being wrong, amplifying error propagation over time.

Using AI for Coding: Speed vs Control

  • One camp reviews all AI-generated code, treating it like a junior’s PR; another admits they can’t keep up with the volume and barely read it.
  • Reported issues: massive diffs, lack of abstraction, duplicated logic, cryptic PR descriptions, and endless patch cycles.
  • A proposed dichotomy:
    • Let agents mostly own the code, relying heavily on tests/benchmarks and accepting reduced human understanding.
    • Or use AI for small, targeted tasks and keep humans fully responsible for architecture and review.

Second Brain, Obsidian, and AI in Notes

  • Many use Obsidian or similar tools as a “second brain,” but opinions diverge:
    • Supporters value searchable markdown, cross-linking, backups, and the thinking that happens while writing.
    • Skeptics see “second brain” culture as hype, mental hoarding, or procrastination; they prefer paper notebooks or minimal digital notes.
  • Broad agreement that the main benefit of note-taking is the thinking required to write; AI-written notes risk removing that cognitive work.
  • Common strategies: separate vaults or folders for AI vs human notes, explicit “generated by AI” labels, or keeping AI strictly as a query/summarization layer over human-written content.
  • Some like AI for transcriptions and summaries (e.g., lectures) and semantic search, but avoid storing AI-generated text permanently because future models will summarize better.

Cognition, Research, and Long-Term Concerns

  • Several worry that over-reliance on AI shifts thinking from “how do I solve this?” to “how do I prompt the model to solve this?”, weakening deep understanding.
  • In research, suggested use is to let AI surface gaps and possibilities, but ensure the final ideas and reasoning are human-owned.
  • Concerns raised about privacy (AI reading private thoughts) and potential large-scale social engineering if such data becomes accessible.