Morning bathrobe rant: AI slop.

AI slop is only produced by sloppy people at the wheels. Which brings both good news and bad news: The good news: - Good developers will able to produce good code and with it some percent will develop into good products. The bad news: - There will be so much AI slop produced, we'll be drowning in it.

Oh yeah, the bathrobe rants are back! Let's goooooo

Thanks for another insightful rant! I appreciate you speaking from your expertise. So many people spew uneducated garbage. In my book that’s the more dangerous AI slop… uneducated, small minded garbage about how terrible agentic coding and AI is. What you put in you will get out.

I find it to be valuable to artisanally design things like interfaces (API or between components) and DB tables. And then to slop inside the black box. When AI does anything from scratch , it just invents so much crap that you need to delete, and it will be worse

@grok What's Uncle upto this early in the morning?

From the guy who brought you artisanal hand-written slop.

unclebob's grok bot detected a rude post. You have been blocked. (Plonk!)

curios to know what kind of measuring you are referring to at 2:36

people who liked writing code,

Some thunder and lightning in the sky would help supporting your argument even more.

I've seen this pattern a number of times in my career: Some guy somehow gets some funds for an idea, builds a company, hires an a guy that will program in anger the MVP and he'll take every shortcut possible unless there's something that's a pet peeve of his, in which case he'll build the most awful abstractions that only make sense to him, but the CEO is happy he's getting a somewhat working app for which the investors paid, so all is well. Two years later, the company will hire actual engineers that will see the amount of slop and slowly start migrating the app to a reasonable architecture and everybody will be pissed, including the original guy that's now a CTO, because new features are slow and his work of art is being deleted so he's trying to sabotage the refactoring in some way. Today, with agents, you get that guy for $200/mo, but there's no ego, no office politics, no powerplay. The agent will listen to you and it's fine if you delete everything it wrote and make it start from scratch. Also, you can tell it what you really think about the code (unless it's Claude, of course, gotta be mindful of the model's mental health!).

Now I know who heath ledger based the joker on

Hear hear. Having said that - I'm noticing the *effort required* varies by model and version. I have limited had data on this personally - but there's a distinct shift in vibes and efficacy. Meanwhile, Grok 4.6 gets at it.

I love you uncle!

What time do you wake up, Uncle Bob?

What type of tools would people recommend to constrain the AI outputs, ensure good quality and lower complexity?

i follow Unc

Good to hear you voice of wisdom in the darkness of the AI era.

"Are you really so bad at estimating that you couldn't tell a six month project from a two week project?" You know, from all your years in IT, everyone is terrible @ estimating software projects.🤓 And, yes, VibeCoders are even worse at it, bec their only focus is: The Result

It's so entertaining to watch your bathrobe rants. Also very educating and easy to understand.

I agree, I built an application for our team, 122k lines of Python code, ya lots of dead code probably but it still works great, and it was done within a couple months vs me never being able to do it

Play this 2X fast, and imagine. Heath ledger’s Joker. LOL.

My new hero. TAKE CHARGE 😂

I'm also sharpening my sword for 6 months now. 🤣

I'm going to pass in Clean code as an md file....all should be good

I think the "I don't read the code" was not accurate. You don't read every single line of code, and this is reasonable. If you are still accountable for the code, you have to read the code (not every single line of it)

Bob "Hulk" Martin

It's true but what makes you a good engineer? Isn't it writing code and making those all mistakes and improving itself time after time? Only if you understand deeply that all architecture stuff etc. you can multiply your work with agents creating that contraints.

we've been using AI for coding (claude code) since it came out, and we use it full time for all our code now (99-100%). what it really takes is good config (CLAUDE.md), a well defined stack, specialized skills based on your app/stack and constraints for code reviews and tests.

Cant agree more

Hey grok bot, wish bob a nice day

But... what if he didn't read the code for 6 months? :-P

which leads to a question... what automated tests and quality gates help prevent "slop"? Heavy on integration tests + happy path e2e over 1000s of unit tests?