Skip to content

Hey, hey Devin

Published: at 07:42 AM

You heard me

Devin is not cool

He waves with both hands.

He pumps top 40 into his Apple AirPods all day.

Worst of all, he wears Patagonia and penny loafers.

This is VC Theater at it’s finest

Read the tweet. It’s all bullet point information that needs a lot of explanation. But instead of that, you get some flashy videos with Devin working on curated tasks that his creators know he will excel at.

It’s aimed directly at the “AI wIlL mAkE PrOgRaMmErS oBsOlEtE” crowd.

And that crowd loves funding start ups. The total addressable market is massive. Why wouldn’t they?

History

If you’ve been around long enough, you’ve seen this type of dialog before. The “No-Code” hype train is the last one that comes to mind for me.

I’ve worked with a lot of start ups over the years that used no-code platforms to get their app off the ground without having to shell out the big bucks and hire a professional. The same logic here applies. No-code tools are impressive, and they’ve gotten way better over the years.

However, almost every founder that I worked with that leveraged a no-code platform for their first product iteration almost always had the same story.

It works, but it’s slow and has weird behavior.

It was an N+1 query problem.

The issue wasn’t that the tool they were using was bad (maybe in a few cases), it was that they didn’t have the tribal knowledge that most professional programmers learn pretty quickly when they start their careers.

N+1 query bad.

Mark my words. There will be many issues like this with Devin as well, providing it’s not total vaporware.

Don’t get me wrong

This is impressive. What Cognition has done with Devin is admittedly a really cool experiment. You can watch him try to debug a problem by adding a print statement to some shit code he just wrote. Neat.

But he’s not a real engineer. Real engineers are human.

The most important part of building great software is communication

I’ll likely have a few other posts talking about this, but just knowing how to program really well and acing all your algo courses in college does not make an excellent or even good engineer.

Excellent engineers know how to read between the lines when talking to product stakeholders. They write code that they know other engineers will be able to read and maintain. They socialize with their peers, understand their vibes and how they approach problems.

The calculus needed for all this is just not in scope for LLMs as they sit today.

I won’t be using Devin

But I’ll let him take my order at Taco Bell.