Ask HN: How do you get into a flow state when using AI to code?

Coders say AI kills the vibe, turns creative work into babysitting

TLDR: The big takeaway: many programmers say using AI to help write software makes it harder, not easier, to get into a deep focus groove because too much time is spent waiting and fixing mistakes. In the comments, people roasted the experience as glorified babysitting, though a few argued it simply creates shorter bursts of productivity instead.

A deceptively simple question on Hacker News — how do you reach that magical locked-in "flow state" while using artificial intelligence to help write code — turned into a full-on group therapy session. The original poster basically said: you don’t. Their complaint was painfully relatable even to non-coders: instead of getting lost in the work, you type a request, sit there waiting, then clean up weird mistakes from a bot that was supposed to help. Not exactly artistic bliss.

And the comments? Brutal. One of the strongest reactions compared AI tools to managing junior workers, except meaner: at least human beginners learn over time. Another commenter delivered the joke of the thread with a dry little dagger: "Welcome to management!!!" Suddenly, the dream of futuristic productivity sounded a lot like answering emails and checking someone else’s homework.

But not everyone was fully doom-and-gloom. A few people argued the old, deep, hours-long trance of focused work is being replaced by lots of tiny bursts instead. That sparked the real tension: is AI ruining creative concentration, or just reshaping it into something shorter, faster, and frankly more chaotic? One especially savage commenter went full life-coach and suggested your unhappy brain might be sending a message. Ouch.

The vibe of the crowd was clear: AI may be fast, but for many, it also feels like interrupting your thoughts for a living.

Key Points

  • The post says agentic AI coding workflows interrupt developer flow because they require repeated prompting, waiting, and review.
  • Faster models are described as improving responsiveness but also increasing the need to fix low-quality outputs, which disrupts concentration.
  • LLM-powered autocomplete is presented as closer to a flow-friendly mode, but the author says it is often either too inaccurate or too slow to help.
  • The article recommends structuring work into well-specified, reviewable task chunks and maintaining detailed handoffs through prompts and AGENTS.md instructions.
  • The author advises keeping software buildable, running quick checks such as cargo check after each change, and regularly reassessing progress in the agentic loop.

Hottest takes

"managing junior AI developers who won’t even grow" — afavour
"Welcome to management!!!" — meetingthrower
"Do something better with your life" — ballooney
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.