People call it "vibe coding" when you let GitHub Copilot, Cursor, or Claude Code shotgun code into your editor and mash accept. That criticism only sticks if you also choose not to read what landed in git diff. The moment you pause, check the generated code in the repo, and gate it like any other pull request, you are still practicing deliberate engineering, you just outsourced the boilerplate typing.

Where the term came from

Here's the Andrej Karpathy post that minted the phrase:

There's a new kind of coding I call "vibe coding", where you fully give in to the vibes, embrace exponentials, and forget that the code even exists... It's not too bad for throwaway weekend projects, but still quite amusing. I'm building a project or webapp, but it's not really coding - I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works.

— Andrej Karpathy (@karpathy) View on X

Best of both worlds

Letting AI handle the first draft removes the grunt work, but reviewing the diff keeps you honest. The diff is where bugs surface, where naming and architecture problems are visible, and where you decide whether to keep or bin the suggestion. Accepting code after that review is no different from merging a teammate's patch: the authorship changed, not the standard.

That distinction matters for businesses looking at custom software modernisation, internal workflow tools, or AI-assisted feature delivery. Fast drafts are useful. Trustworthy delivery still depends on review discipline, environment control, test coverage, and someone who understands the operational consequences of getting it wrong.

Double code review via commit messages

There's another hidden review loop: tools like GitHub Desktop or GitHub Copilot can auto-generate commit messages. When you let them summarize the change, you instantly see whether the diff tells a coherent story. If the generated summary is wrong, you probably missed a logical wrinkle in the code itself. It's a second pass, because now both the code and the prose description need to make sense.

And now GitHub Copilot can also review pull requests directly in GitHub. In practice, that's genuinely valuable: it catches edge cases and consistency issues humans often skip in quick reviews. It can be a bit nitpicky at times, but I'd rather filter noisy comments than miss a real defect.

Practical workflow

• Use Copilot/Cursor/Claude Code to draft code, but never skip opening git status or your IDE diff view.
• Edit the generated code until you're confident you'd approve it if it arrived from a colleague.
• Run tests or linters exactly as you would for hand-written work.
• Let GitHub Desktop, Copilot, or Claude Code suggest a commit summary, then tweak it—if the summary feels off, re-open the code and figure out what you missed.

That workflow gives you the best of both worlds: speed from AI drafting plus rigor from human review, with an extra safety net provided by AI summaries of your commits. That's not vibe coding. That's just modern software development.