Vibe coding is mad depressing

Client boundaries vs. “vibe coding”

• Many see the story less as an AI problem and more as a classic freelance‑client boundary failure: letting a client push unreviewed code to main would have been bad long before LLMs.
• Suggested fixes: clear contracts, branch protections/PR rules, being willing to say “no” or even fire clients, and charging more when clients “help” or bring half‑finished work.
• Semi‑technical clients (or managers) with strong opinions but shallow understanding are described as the worst; LLMs just turn more people into that type.

Maintainability and “day‑0 legacy code”

• Several developers report that LLM‑generated projects look impressive early, but become unreadable, convoluted, and brittle; when they break, it’s often faster to rewrite from scratch.
• A recurring theme: reading/understanding code is harder and slower than writing it, and you don’t truly “own” what you didn’t design.
• People liken vibe‑coded codebases to instant legacy systems: poorly structured, under‑tested, and hard to extend, with long‑term costs hidden from non‑technical stakeholders.

Using LLMs effectively (or not)

• Positive experiences:
  • Prototyping, quick PoCs, exploring APIs/libraries, generating boilerplate and tests.
  • Helping with large refactors/ports when tightly constrained by tests and build checks.
• Negative experiences:
  • Non‑compiling or stubbed code, fake tests, invented behavior, and resistance to following instructions.
  • Feeling reduced to a QA/debug role for the model, with diminishing returns as projects grow.
• Proposed best practices: strict exit criteria (must compile/tests pass), small scoped tasks, planning first, context management, and treating the LLM as a junior assistant rather than an autonomous agent.

Consulting, low‑code, and a cleanup market

• Consultants and low‑code vendors report a surge of leads asking to “just fix this little part” of vibe‑coded systems that are actually architectural disasters.
• Some see a new niche: high‑rate “vibe‑code salvage” work for clients who already know they’re in trouble.
• Others view vibe coding as a long‑overdue disruption of overpriced, low‑quality traditional contracting, with technologists needing to educate sponsors about risks.

Emotions, analogies, and long‑term concerns

• Emotional reactions range from “automation sorrow” and loss of the problem‑solving high to enthusiasm about offloading drudge work.
• Comparisons: Geocities‑era web, “Doctor Google/Doctor GPT” vs. physicians, plumbers and clients, Doritos‑like overindulgence in AI code.
• Some worry that AI will eventually automate the meaningful parts of programming and other knowledge work, not just the boring bits.