All About Vibes

1. Hackerman

Oh boy do I have a bone to pick with the computer nerd trope.

Since the dawn of computers, the elusive computer nerd has always been a recognizable trope. Glasses, clip-on tie, half sleeves shirts, and an inability to talk to women — all tell-tale signs of the trope. I still see this guy whenever I head on over to the cinema, and it takes me out of the movie each time.

Of course, there was a brief period of time when The Matrix came on to the scene and made hackers in full leather seem cool, but that’s something you and I are better off talking about on DMs.

Oh yeah, and lets not forget how groups like Anonymous and shows like Mr. Robot also popularized the hacker in the hoodie typing away in the middle of the night with not a care of the world.

With each depiction of the computer expert, the act of software work is shown as something almost other worldly — where mind meets flesh with the rapid clicks and ticks of the keyboard. Although these characters do get dunked on, their skills are seen as impossible to replicate.

But you and I both know that this idea has never been more wrong than it is today.

2. Sign of the Times

What’s different? AI, that’s what.

I still remember when I was nearing the end of my CS degree (yes, I’m technically a computer scientist!) that everyone was hyped about GPT being able to cook up front-ends like it was nothing.

Twitter would be crawling with showcases of the model generating websites in response to mere prompts.

Excellent demos? Yes. Achievable back then at scale? Not really.

My buddy Saad (now my CTO) and I would spend sleepless nights getting our final project to function.

To some extent, we felt like circus animals jumping through a ringmaster’s hoops.

We were in the car last week reminiscing about the times when he made a pretty telling statement — “If we had GPT back then, man would I have gotten more than 2 hours of sleep.”

When Antematter started, it was just a few of us in our college’s cafe mashing away our keyboards.

This was back when we were making contributions to Solana.

Today? The dynamic’s completely different — almost all the engineers on my staff get decent hours of sleep and don’t click away like they do in the movies.

That’s the advantage afforded by vibe coding.

3. Good Vibes

So, what exactly is ‘vibe coding’?

Well, might as well let the father of the term explain it himself.

Too much? I’ll summarize.

Vibe coding is the idea that you stop thinking about code — period.

Instead, you get your AI assisted tool to write everything for you. All you have to do is tell it what you want.

You don’t think about what the LLM wrote — you take it at face value and plug it in immediately.

Something’s off? Just slap the error back at your tool, word-for-word.

Tools like Cursor have made coding so easy that anyone can do it — even my business’ chief memes officer can write like a pro with GPT at the steering wheel.

I met my 12 y/o niece at a family dinner just recently, and she showed me a fully functioning Roblox game she had built in the span of two weeks.

She didn’t do anything — she just followed some tutorials on how to get GPT to make Roblox games.

My own CTO now occasionally throws what he calls “donkey work” at GPT so that he doesn’t have to bat an eye.

AI-assisted coding has already become part-and-parcel of Antematter, and it might very well change into a no-code environment when the tools get better.

Think this is temporary? Guess again — Y Combinator’s already reporting that more than a quarter of their recent batch had 95% of their codebases written by AI.

And it’s only going up from there!

4. We Need a Little Controversy

But of course, not all that glitters is gold.

There’s more than a fair share of issues with getting AI to do your codebases for you.

For starters, vibe coding dictates complete abandonment of concern for code — a vibe coder relies solely on the vibes they’re getting from the AI and software.

This, of course, is incredibly risky when you’re working on production-ready code that’s about to be used elsewhere.

I’m forced to keep a man in the loop for any project that’s heavily tied to a framework of rules — I’d only do the client harm if the AI messed up by even the slightest bit!

If you start reviewing and testing the code out yourself, you’ve eliminated the point of vibe coding — you’ve, in Simon Willinson’s words, only used the AI as a typing assistant.

An NYT journalist — Kevin Roose — tried pulling off a stunt where he relied solely on vibes to build a couple of web applications.

Roose, of course, noted that AI generated code was quite error prone. And it was readily pointed out by folks like Gary Marcus that Roose’s AI had worked with data of this sort before.

Heck, might as well make room for another Antematter anecdote.

Just last week, our research scientist and an AI engineer teamed up to interview some prospective senior-level AI engineers and data scientists.

To their shock and horror, there were several candidates who could talk the talk but couldn’t walk the walk — they didn’t know a lick of python, despite working on AI with python!

5. Dissipation

I’ve had this argument with some of the folks within my own business, and we’ve kind of come to a mutual understanding.

As our research scientist puts it, “The tools of the trade change, but fundamentals are forever.”

That’s certainly a sentiment my CTO and I are willing to stand by.

The tools might be getting smarter and writing up more robust code, but removing human thought from the equation just seems to be a bit of a stretch.

Perhaps Willinson’s idea might be right after all — we need AI as typists.

If you’re interested in AI agents for your organization or products, visit our website at antematter.io.