With AI you need to think bigger

Original Article ↗ Hacker News Discussion ↗

AI as an Ambition Multiplier

  • Many commenters report a dramatic drop in “activation energy” for projects: tasks once shelved as “too big” or “too fiddly” (ERP migration, ML on Raspberry Pi, custom scanners/bridges, whiteboard tools, GUIs, webhook services) now feel achievable in hours instead of days or weeks.
  • Especially strong impact on:
    • Glue work (joining APIs, wiring services, scripts, Obsidian plugins, deployment boilerplate).
    • Frontend/CSS/UI for backenders.
    • Porting between languages/frameworks and refactoring smaller units.
  • AI is likened to earlier shifts (compilers, internet, cloud): it lets you build more with less, and soon others will expect you to.

Where LLMs Work Well vs. Poorly

  • Strong for:
    • Learning new domains using simple language and good abstractions.
    • Prototyping, scaffolding, rote boilerplate, small targeted changes, test generation.
    • Rubber-duck debugging and exploring multiple approaches quickly.
  • Weak for:
    • Novel or under-documented problems; niche libraries with little training data.
    • Large, messy, evolving codebases and long-term maintenance.
    • Deep design/architecture and creative solutions that deviate from the status quo.
  • Several note drives “littered with failed AI projects”; others find AI code “close but not complete,” still requiring deep manual work.

Skepticism, Hype, and Risk

  • Pushback against claims like “you can now do literally anything”: people report hallucinations, subtle bugs, wrong libraries, and AI reintroducing toy-example patterns into serious systems.
  • Concern about overconfident marketing, low-quality benchmarks, and “gaslighting” about current capabilities.
  • Some view AI as mostly surfacing the same solutions that used to be found via high-quality search, just faster and with code extracted.

Careers, Skills, and Learning

  • Senior engineers describe both excitement (10x leverage, boredom cured) and anxiety (fear of obsolescence, shrinking junior roles).
  • Counterarguments: conceptual depth, system thinking, and domain knowledge make AI more powerful rather than replacing experts; AI can’t yet decide what to build.
  • Speculation that non-coding roles (PMs, “software anthropologists”) might increasingly drive “vibe programming,” but others argue translation of business needs and long-term maintenance still demand engineers.
  • Teaching: early anecdotes that students using AI tools both deliver more and grasp fundamentals better, but worries remain about over-reliance and deteriorating skills.

“Think Bigger” — With Caveats

  • Consensus that AI makes more ambitious personal and professional projects feel approachable and often practically doable.
  • Several warn that complexity, invariants, and long-term evolution remain hard; AI accelerates entry and expansion but not escape from poorly designed systems.
  • Recommended stance: embrace AI for speed and scope, but stay critical, verify outputs, and treat it like a powerful but fallible junior collaborator.