RunEvr
Agentic project management environment for creatives
Start new thread

🚩 Why are AI-generated texts flagged but code isn’t?

by
•

There’s a growing tendency to flag AI-generated text. There’s even a term for low-quality AI-generated content - AI slop.

And honestly, I understand why.

Communication should feel real. If every comment, post, or message is generated by AI, the picture becomes blurry.

But then I have a question.

AI is now generating huge amounts of code, UI, documentation, copy, and even product ideas.

And when AI generates the code behind a product, there can be real issues too - security, bugs, hidden dependencies, or simply code that nobody fully understands.

There are also dozens of memes among developers about the classic rule:

“If the code works, but you have no idea how… don't touch it.” =D

Yet we don't usually see a product marked:

This product was built with AI-generated code.

So why is AI-generated text a problem, while AI-generated code is mostly accepted?

Is that fair? =D

What do you think?

125 views

Add a comment

Replies

Best

I think one of the big differences between the two, is code-based AI changes are often gated through code review. Most source code platforms offering PR-style workflows, allow the workflow to involve a human-reviewer that often validates, or rejects code generated because of these issues. Also, because of automated tools like vulnerability scanners, etc it generally means the quality of the integrity of the code is superior. Ultimately of course, code has to compile, it has to test well, these are further gates that signify the quality of the generated code.

AI based free text is a different issue - are there platforms and tools out there that allow the same fidelity of gated checks that code-based offers? If not, honestly maybe that could be a niche product right there?

 I guess there are some tools, but I don’t think they work perfectly.

Today, more and more text-related products are training AI to generate more human like speech =) So everything is getting mixed up...

And regarding high quality code - no one said that AI-generated text isn't high quality too =)

 Certainly not implying AI text is low quality, but code has such a high bar of quality tools that it is not a fair comparison I don't think. Happy to learn about new tools people find though!

Code gets questioned too, it just happens inside companies rather than in public. The bigger the place, the louder it gets, and what people are really asking is who has to answer for it when it breaks.

I use AI for all of my coding. What I do about that is test, and do the security work properly. The task manager has 131 Playwright spec files and 275 tests, and one of the dull ones caught a real hole last week: a key that should have stopped working still worked, because the revoke was sitting inside a transaction that threw right after it.

Quality and security are both better now than when I wrote the code myself, which I didn't expect. So for me it doesn't come down to whether AI wrote it.

 This is the case when the person checking the result actually understands everything they’re doing.

The same goes for text: if the person knows what they want to say and uses AI just to help form it, that text will probably never get flagged.

What worries me more are products that seem to work perfectly, but are doing something unwanted underneath.

 Fair worry, and you're not stuck with it. Ask an AI agent like Claude Code or Codex to check whether a service is safe and it'll go through it in no time. Be careful doing that against a SaaS though, since poking at someone's servers could look like an attack. For an app you install on your own machine there's no problem.

 Definitely don’t want to be considered an attacker =D

Thanks for the advice!

I think the difference is that text is communication, while code is implementation.

If AI writes my comment, it can feel like I outsourced my own voice. If AI writes part of my code, I can still review it, test it, understand it and take responsibility for the result.

That said, it is funny that we’re often more suspicious of an AI-written paragraph than thousands of lines of AI-generated code.

For me the real problem isn’t AI-generated anything. It’s using AI without judgment.

 Funny moment indeed =)

And I agree that the problem is judgment. Not just judgment itself, but someone who can actually judge the result.

It can apply to design, code, text, photos - everything.

If AI can write both the text and the code, why do we judge the output differently?

 It can also take photos, shoot videos, and much more =D

I think the difference comes down to expectations. When we read a comment, we expect a genuine human perspective. When we use a product, we expect it to work reliably and securely, regardless of how the code was written. =D

 Wise answer ! =) But there might be some security factors - but it's my own opinion.

When AI writes my code, it copies the style and conventions of my codebase and looks pretty similar to what I'd write. You also don't need creative original ideas for most code.

When AI writes a post/reply on social media, it has the worst linkedin style take imaginable.

 Agree about the LinkedIn style =D

And about the code - yes, in your case you have your own codebase and deep understanding.

I’m more thinking about people who create things with AI without much technical knowledge.