← Back to blogYouTube Video
Published June 18, 2026
"Who cares what the code looks like?" is dumb
Can't play the video or having issues? Here's the direct link.
Summary
I argue that “if it works, who cares what the code looks like” is a dangerous standard for agentic coding because it assumes the agent perfectly understands the codebase, has complete context, and won’t hallucinate bad code—none of which usually holds. Agents often have a weak understanding of the codebase, make bad assumptions when context is missing, and as statistical models will eventually produce buggy or hallucinated code. My take is simple: code quality still matters, and reviewing the agent’s output is still essential.
Key takeaways
- Review every agent-generated change before trusting it; working output can still hide buggy or hallucinated code.
- Give agents complete context about the codebase, because missing context forces them to make bad assumptions.
- Treat agents as statistical models that will eventually produce bad code, so don’t abandon code quality just because it runs.
Description
Just because AI models can write thousands of lines of code for you doesn't mean that all of it is correct. Who the heck came up with the idea that we can just ignore the code that's being written?
Transcript
The problem with the “if it works, who cares what the code looks like” crowd is that, for agentic work at least, you assume that, one, the agent has a perfect understanding of how everything works. And two, you've provided every single piece of context and left nothing to the agent's assumption. And three, the agent won't hallucinate bad code.
And usually, none of those things are true. Agents usually have a less-than-ideal understanding of the codebase. They make bad assumptions when you don't give them enough context. And I can guarantee you, 99% of you folks are not giving enough context to your agents, which is why they are forced to make assumptions about the codebase they're working in and how they should write new code.
And all of this is eclipsed by the fact that agents are statistical models, and they will hallucinate bad code and buggy code sooner or later. So no, code quality is still important. Reviewing the agent's output is still very important, and you need to make sure you're doing all of that.
Dude, you’re at the bottom of my landing page.
Just book a call already if you’re that interested.
You scrolled all the way here.