AI coding made me faster, but I can't code to music anymore

Where AI Helps and How It’s Used

• Strong gains reported for:
  • CRUD / API glue, boilerplate, schema/test generation, migrations.
  • Data pipelines, format conversions, scrapers, small internal tools.
  • Wiring auth, infra/config following wikis and logs.
• Works best when:
  • The human does the reasoning/architecture and uses AI for implementation details.
  • Tasks are “common” and well-trodden; AI struggles in poorly understood domains.
• Described as a “force multiplier” for experienced devs who know what they want but don’t want to read every doc/manpage.

Perceived Productivity vs Actual Velocity

• Some claim order-of-magnitude speedups and “team of interns” vibes.
• Others find:
  • Long QA/debug cycles negate speed.
  • Agent workflows can produce large broken patches that get scrapped.
  • They become overall slower but end up with higher-quality designs.

Cognitive Load, Flow, and Role Shift

• Many say AI sessions are more cognitively intense:
  • Constant prompt → code-review → re-prompt loops.
  • Less “typing trance,” more high-level planning, specification, and evaluation.
• Feels like being an engineering manager or senior reviewing a junior’s work, without the managerial rewards.
• Debate over “flow”:
  • Some insist true programmer flow (entire program in head, near-error-free typing) is incompatible with LLMs.
  • Others say they still experience flow, just with different rhythms and more multitasking.

Music, Attention, and Individual Differences

• A recurring theme: AI coding makes it harder to listen to music, especially with lyrics.
• Explanations offered:
  • Prompting/reviewing competes with language centers used for listening.
  • When AI does the “manual” part, what’s left is pure thinking, which doesn’t mix with music for many.
• Others report:
  • Instrumental/techno/metal still works fine.
  • Music can either aid focus or be a major distraction, highly individual.

Quality, Tech Debt, and Debugging

• Concerns:
  • “Slop” code, subtle bugs, increased tech debt, harder debugging, and loss of institutional knowledge.
• Counterpoints:
  • LLMs make refactoring and cleanup cheaper if you understand and constrain the code.
  • Heavy emphasis on tests, tools (e.g., Playwright agents), and strict version control.

Work, Enjoyment, and Power Dynamics

• Many fear:
  • Short-term joy of “a day’s work in an hour” will become a new baseline for output.
  • Less enjoyment, more alienation: coding turns into supervising machines rather than crafting.
• Some argue this mirrors historical automation: productivity gains likely accrue to employers, not workers, unless resisted collectively.