Hacker News front pageJul 24, 2026, 18:12 UTC+8

How Do We Stop Vibe Coding?

Alex Klos argues vibe coding erodes a developer's understanding of and trust in their codebase, and all current solutions are immature. He cites Grady Booch's comparison of AI coding agents to compilers, framing the shift as moving from writing code to expressing intent. Right now, it feels like pulling a slot machine lever, churning out unreliable code. The post evaluates Markdown specs, Skills, Spec Kit, Kiro, TDD, and others, concluding none yet solve the core trust problem.

Why it matters: A substantive dev opinion piece that systematically examines the vibe coding problem and immature solutions, with a nice Booch compiler analogy. Docked because it's a personal blog without first-party data or experiments, and some tools cited (Kiro, CodeSpeak) are obscure, lim...

Read the original ↗Export Markdown

Take

Alex Klos's piece is worth reading not because he has the answer, but because he walks through every current fix and explains why it falls short. Markdown specs are too rigid. Skills are too granular. Spec Kit and similar toolchains add too much ceremony. Kiro is early. TDD sounds right in theory, but in practice nobody is genuinely writing tests first with an AI agent. The Grady Booch quote he pulls does the real work here: AI coding agents are like the arrival of compilers. We're moving from writing code to expressing intent. That framing is sharper than the slot-machine metaphor.

I'd discount this about 30%. Klos admits he hasn't typed a line of code since late 2025, so this is a heavy agent user's internal reckoning, not a neutral audit. CodeSpeak and Scryer get name-dropped but not unpacked—treat those as leads, not conclusions. The one thing he nails: there's no retreat. The question isn't how to stop vibe coding. It's how to make the intent-to-code pipeline verifiable.

Source: Hacker News front page · alexklos.ca
Published: Jul 24, 2026, 18:12 UTC+8 · Jul 24, 2026