Robinhood’s CEO Says Majority of Its New Code Is AI-Generated

“When asked what percentage of Robinhood’s new code is AI-generated, Tenev said that it was around 50%. That’s a higher percentage than what Microsoft and Google have previously said, with CEOs Satya Nadella and Sundar Pichai estimating around 30%.

“We’ve moved from GitHub Copilot, which is an autocomplete system, to Cursor, and now things like Windsurf, where nearly all of the code is written by AI,” Tenev said. “It’s hard to even determine what the human-generated code is.”

Tenev said AI had contributed to the company’s cost and efficiency stabilization, he said.

“We have been pretty mum on how much we’ve been using it internally,”

**Vibe coding = Vulnerability as a Service**

In the next couple of years that platform is going to be full of security holes and performance issues.

This explains why I suddenly own shares of Blockbuster and Circuit City.

As a programmer, I find this doubtful. Our org has so far determined AI code is at best a break even in terms of productivity, at worst we spend way more time fixing it than if we just wrote it ourselves.

They deserve to drop 20% in value announcing something like that

They’ve entirely replaced autocomplete for code with AI, and every time you use the autocomplete it counts as “ai generated”. Even if that “ai generated” content would have been exactly the same as the autocomplete before.

Hugely inflated these numbers everywhere.

Not to say there isn’t AI generated code and vulnerability as a service as others have said, just that this swap of old autocomplete vs gen ai autocomplete hugely inflates these numbers.

I’d love to know how much code was generated by standard autocomplete before, the autocomplete for code was significantly advanced and I often miss it over the AI slop autocomplete we get now.

We are one step closer to Boeing announcing their code is mostly AI generated and the FAA saying their inspections are conducted by AI using code from Big Balls. Good luck to us all.

So i just spent a week doing a deep ai experiment with 6 systems.

14 hours each day.

The only thing ai improved in code generation is it took away a lot of typing code time away. The ai all made huge mistakes that had to be fixed by repeated back and forward they all hallucinate hugly om code that doesn’t exist.

One of the main issues is, if they decided that using a library function was the answer they put it in the code they would then become absorbed by using that function even when told to remove it.

I mean the AI was awesome for very very specific tasks.

But your not going to survive long on new code building with just ai. You still need someone who knows exactly what the steps are.

Also AI do a lot of extra lines most developers dont so if you look at line counts produced as your metric, it’s going to be inflated.

Ai isn’t ai right now. It more of a programming partner reminds me of the early 2000 when companies did partner development

I worked at a big tech company. If I say “I’m eating chicken noodle” and you say “soup!” then you can claim to have written 20% of my sentence.

Just sharing in case that makes anyone think differently about these stats.