enterprise vibe coding
August 4, 2026 ยท 2 min read
I've been thinking about writing about the somewhat alarming emergence of vibe coding in the enterprise for quite some time. Trying to write something on the topic that didn't immediately sound derivative or boring was always going to be the challenging part.
I guess what surprises me the most is how quickly we arrived here. I remember playing with Github Copilot while it was still in technical preview during lockdown in late 2021. It didn't yet have the now standard chat interface we have come to know and love and instead you prompted it by writing what you wanted via comments in existing files and hoped it did what you asked.
Fast foward 5 years and we now have fully agentic AI-first editors such as Cursor and advanced CLIs such as Claude Code that live in your terminal and can be left to run for hours or even days without intervention. These tools are fantastic and capable of producing entire applications not just small refactors or those unit tests that you hate.
Unfortunately as with most things there can be abuse and lately I'm starting to see the same problem occur repeatedly. Engineer gets given ticket, they vibe code their solution, their MR gets accepted, the codebase gets considerably worse.
We have always had to mitigate low quality code whether that be from a junior with the best of intentions or from a contractor looking to get their job done in the quickest dirtiest way possible. What stands out to me this time is the sheer volume of code that can now be produced without much thought and the kind of engineers we are becoming.
Old school rules apply:
The best form of offense is good defence. Ensure necessary testing is enforced at pipeline level and leverage the latest advancements in testing such as browser mode, visual regression testing etc.
Reviewing code is part of the job. Instilling in fellow teammates the importance of taking the time to perform a thorough review and the wonders it will do for their skills as an engineer.
Document acceptance criteria in Storybook. Something I've been doing lately is ensuring my team has Storybook stories for each piece of acceptance criteria. It's early days but it does seem to be working and ensuring functionality doesn't break.