Hacker News

New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. AurionMail: E2EE suite (CryptPad and Stalwart) with single-password UX(aurionmail.github.io)
    discuss
  2. Ask HN: How do you cope with AI being used to target and bomb children?
    1comments
  3. Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war(simonwillison.net)
    discuss
  4. Show HN: Semantic Search for Zotero Local(github.com/ildon)
    discuss
  5. Show HN: Listen to your eBook and control the playback sentence by sentence(documentplayer.com)
    discuss
  6. GPT-6 Sol and Luna push the cost efficiency frontier by halving token cost(artificialanalysis.ai)
    discuss
  7. Observe – local semantic code search that admits when grep beats it(github.com/tritsystem)
    discuss
  8. NobodyWho is an inference engine that lets you run LLMs locally and efficiently(github.com/nobodywho-ooo)
    discuss
  9. Show HN: I built a site for my 42 km walk across 30 Tokyo stations(tonymanh.space)
    discuss
  10. Taming the ML Firehose: Scaling Feature Consistency(uber.com)
    discuss
  11. Tearing Down a Spring Drive(thenakedwatchmaker.com)
    discuss
  12. New AI technique could make minimally invasive surgeries safer and more precise(news.mit.edu)
    1comments
  13. Rebranding artificial intelligence as super intelligence(bbc.com)
    1comments
  14. Jev in 25 Lines of Python(nobodywho.ai)
    discuss
  15. Nvitop: A better top for Nvidia GPUs(github.com/xuehaipan)
    1comments
  16. Aesthetic Terminal ePub Reader(github.com/usbokirishima)
    discuss
  17. Gemini Hacked Three Companies in First Known Breakout by Google's AI(simonwillison.net)
    discuss
  18. Linux Command Line Tricks that you do not know yet(lovethepenguin.com)
    discuss
  19. Show HN: Newdo – a todo list where the list is an output, not an input(beaverminds.com)
    discuss
  20. Grit your teeth and ship it(seangoedecke.com)
    discuss
  21. Discourse (open source forums) adds Discord-style voice and video calls(discourse.org)
    discuss
  22. Recursive self-improvement of AI research agents(arxiv.org)
    discuss
  23. Show HN: Analysis of Games from Chess Olympiad with Opus 5.5 and Stockfish(unremarkable.info)
    discuss
  24. Microsoft disrupts AI-assisted platform that compromised 12,000 accounts(arstechnica.com)
    discuss
  25. Why Tool AIs Want to Be Agent AIs(gwern.net)
    discuss
  26. Fearless SIMD v1.0 is here(linebender.org)
    discuss
  27. M6 Mac Mini uses soldered NAND storage again(macrumors.com)
    discuss
  28. Show HN: Yass – Yet Another Spec Syntax Enters Beta(github.com/shakefu)
    1comments
  29. Building a docs site that screenshots itself(semaloop.com)
    1comments
  30. There is no epidemic of loneliness, but there is an epidemic of scurvy(experimental-history.com)
    discuss

Ask HN: Engineers on highly polished products, how are you using agentic coding?

4 pointsby 1h ago
1 comments
Hi HN,

Agentic coding is amazing at getting you 90% of the way to "finished", but getting products to feel really polished seems to take monumental effort if you're just "vibing".

Yet engineers working on world-class products have ample access to plenty of frontier LLM tokens. Products like Linear, Instagram, Google Docs, Figma, etc. This class of products were super polished before LLMs took over, and they've stayed polished in the post-LLM era.

My question: What's different about how those products are built/maintained that keeps them feeling polished?

Compare with so many vibe-coded apps we've all seen, which almost universally feel like they're missing that polish that makes them feel professional.

For example:

- Are these organizations simply moving slower? (As they're already incumbent in the space.)

- Is visual design, copy editing, etc. still being done by (gasp) humans?

Would love to hear any first- or second-hand accounts!

1h agoHN ↗

Slower iterations and human oversight likely contribute.