My bets for what to learn next ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌
View in Web Browser

 

Develop Yourself

Avoid Extinction

"Most of the people writing in JavaScript are not programmers."

Douglas Crockford, 2001

Swap "JavaScript" for "Claude Code" and you could post that on X today. You'd get 4,000 likes from guys with anime avatars.

Crockford was defending the language, by the way. He was describing the reputation it had. JavaScript was a toy for amateurs and designers. Grown-up programmers wrote C++ or Java.

Then JavaScript became one of the most popular languages on earth. Companies hired for it by the thousands. Laugh at it all you want (1 + [] is a real knee-slapper). It paid a lot of mortgages, including mine.

The pride problem

A lot of developers wrap their identity in being smart. So when normies get into the profession, they can get weird. They draw lines, with the TRUE programmers on one side and everyone else on the other.

Arguments about which language counts as programming are stupid. If you get into them, I think you're even stupider.

Last week I saw someone on X ask how many vibe coders could build a compiler, a transpiler, or Docker.

Why the f*ck would they build that? They're busy trying to make money.

The worst advice I could give you

In 2019, if you told me you were going all in on Go or Rust, I'd have said great idea. Become the expert. Own a niche. Have leverage.

Today that's the single worst advice I could give you. LLMs will beat you at syntax and new languages faster than you can imagine. Don't race them there.

So what can't they beat you at? Here's where I'm placing my bets.

1. Being the human people trust

Most people don't trust LLMs (fair). Someone has to explain what the model produced, why it's right, and how to use it. That person has a job for a while.

2. Teaching

Millions of people are just getting started with these tools. Somebody has to show them how to use them well.

3. AI engineering

This means building around the models, not training them (please, no). Systems and products that rely on LLMs and that people will pay for. That's where the money is.

4. Cleaning up the slop

People are shipping vibe-coded junk at an alarming rate, and the bugs are piling up with it. The models are very fallible. More people building means more broken software, which means more demand for people who can fix it. The fixes are usually architectural: patterns that don't scale, code no human can read (which also makes agents burn tokens digging through it), and designs that choke at 10 concurrent users. A non-technical founder won't spot any of that, and they don't want to maintain it either. Software maintenance is about to get sexy again (ok, maybe not sexy).

5. Data

When anyone can build anything, most of what gets built is junk. Good products come from data. Run experiments. Find where users drop off. Then feed all of it to an LLM and ask for patterns and insight. Then... fix that sh*t.

6. Forward deployed engineering

FDE is one of the hottest roles (so fetch) to come out of this AI explosion, and it'll get bigger. Watch for internal FDEs too. Companies will ask whether to hire an expensive consultant or take the engineer who's already automating her own team's workflows and let her teach the rest of the org. I'd bet on that gal.

What I'm NOT betting on

Learning the models is a short-term gain, and shorter than usual. They're getting good fast, and they're learning from how engineers use them. If you're spending your nights building elaborate skill libraries and software factories, I think you're optimizing for a world that's about to disappear. Pretty soon we'll just say "do this, connect it to my tools, and let me steer."

But hey, I've been wrong af more than a few times when it comes to AI.

I think the devs who crush it in 2027 won't be the ones who can write a compiler. I'm betting they'll be the ones who can explain technical concepts to non-technical people, and who leverage AI to experiment and build features around the models.

Build the skills that won't go extinct (too fast at least)

The Parsity AI Engineering program teaches you to build around the models: RAG, agents, evals, and the production systems companies actually pay for. We pair you with a senior mentor so you learn from people who have built the systems we teach.
Check out the program

See you around,

Brian
Unsubscribe | Sent by Parsity
47 S Meridian St • Indianapolis, IN • 46220