The truth is, I don't use those modern AI tools much. I'm not doing much coding and when I do, I've been using Clojure or Elisp - not so mainstream so as to have all sorts of AI support. For writing? I prefer do to it myself. Anything else? Well, to be honest, there's not too much going on in my life where I feel like AI will improve things. Maybe if I were writing important business emails or I was cranking out tons of boilerplate code for a company but as a retiree just working on personal projects, I just don't see the need.

In any event, this newfangled Vibe Coding thing caught my attention.

The idea is that you have AI generate a programmed application from repeated prompts. I'm not going to say you're programming or coding via AI because you're not but more on that in a bit.

You prompt an AI code generator to write something. If it works, great, if not, restate the prompt or give the next prompt to address the issue. Whatever the AI comes up with, accept it. If the solution works, you're done, if not, rinse and repeat.

Apparently the Vibe Coding label and hype got a big push when Andrej Karpathy tweeted this:

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 possible because the LLMs (e.g. Cursor Composer w Sonnet) are getting too good.

Also I just talk to Composer with SuperWhisper so I barely even touch the keyboard. I ask for the dumbest things like “decrease the padding on the sidebar by half” because I’m too lazy to find it. I “Accept All” always, I don’t read the diffs anymore.

When I get error messages I just copy paste them in with no comment, usually that fixes it. The code grows beyond my usual comprehension, I’d have to really read through it for a while.

Sometimes the LLMs can’t fix a bug so I just work around it or ask for random changes until it goes away. 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.

Now, in the past couple of weeks, I've seen a number of examples of vibe coding, testimonials about it and, to me the one I really want to address, people floating that it's something we should be teaching, at times, as a valid coding technique.

Reading the start of Karpathy's tweet, it sounds like he's pretty excited but reading the end we get "It's not too bad for throwaway weekend projects" and "it's not real coding." It appears some people are forgetting those parts.

The first time I saw vibe coding - prior to my knowing the term was when watching a YouTube video. The author was showing how to do something, I forget what and then said he coded up a script to help along the way. Then he said he didn't actually code it but used AI. He then did give the disclaimer that he really didn't know how the script worked and would have been reluctant to use it if the task at hand was actually a really important one.

That raised a big point - with Vibe Coding, the human prompting the AI might have no understanding of the solution. This means they won't be able to customize it and more importantly, won't be able to audit the code to make sure something bad isn't happening.

Do we want a vibe coded banking application? How about a vibe coding insurance claims approval system. Don't even think about a vibe coded surgical assistant. How about military control stuff? I mean the WOPR was bad enough with Falken writing the code. Imagine if it were Vibe Coded.

This means that while vibe coding can be useful for a quick prototype or proof of concept, it's going to be extremely dangerous to actually rely on Vibe Coded solutions.

Of course, this is America where common sense and prudence always takes a back seat to the dollar so I wouldn't be surprised to see attempts at Vibe Coded "solutions." Particularly if it can be set up so that the customer and common man gets screwed by any issues and not the business that rolled out the problematic code.

The other thing I keep seeing with Vibe Code is that the examples people show are variations on well known themes - an ecommerce site, Tetris, or something else familiar with tons of pre-coded solutions out there. This makes sense given the way AI works. It can do a reasonable job at generating variations on a theme - remix the existing stuff. Let's see about Vibe coding something brand new.

In any event, for prototyping I could see a big win but I'm pretty sure Vibe coding will be used in lots of places where it shouldn't. For now, I'm placing Vibe Coding in the same category as other AI coding assists - helpful and maybe even extremely helpful for real developers. Also helpful for non-coders at the prototyping or toy level but I'm not seeing the end of the human software developer in the near future.

But now let's look at the education side.

I've already heard the praises for Vibe Coding on the education side - you can create without knowing a line of code (or even that it is coding) and that it's empowering,

You can create without knowing a line of code, I guess, but create in a way that reminded me of something I think Mitch Resnick said at this past SIGCSE - "Do you want your kid to learn to play Guitar Hero or do you want them to learn to play guitar?" If they Vibe code a solution what are they really learning? Sure, there's a lot of mundane, nitty gritty code stuff that's annoying to teach and learn about coding but that comes along with things like problem solving, dealing with complexity, and yes, even dealing with the details.

Vibe coding, like other AI techniques can be tremendous time savers for people who already know but if they're used too much like a crutch before people know, then they'll never learn.

Look at math. Sure, we use calculators but does that mean there's no value to learning basic numeracy, number facts and operations? Sure, we don't have to do them super fast, nor memorize all the times tables but there was absolute value to learning the basics and only then going to the calculator.

We now have all sorts of mathematical problem solving tools. Does that mean there's no value in learning algebra? While there are many who would say that there are a lot of specifics in an algebra class that maybe isn't so important, anyone being honest with themselves will agree that there's also a lot of fundamental problem solving that's learned through taking algebra.

If someone just learns to Vibe Code, it's just like hiring a bad software engineer - telling them what to build, and when they come back with an error, not being able to help fix it but basically guessing at some other prompt.

So in my opinion, you're not really creating with Vibe Coding - you're asking something else to create for you and it's not empowering for a student because the student isn't growing as a result of using the tool.

Could Vibe Coding be used successfully in an educational environment? Sure. Particularly when it comes to prototyping but that's for later.

For those intro levels, let's not teach Guitar Hero, let's teach guitar.