🤖 Inside the $1.5B AI collapse

Today we're switching things up.

Normally I bring you news, tools, tutorials. Not today.

Today we're talking about the truth nobody's saying about AI and coding. Because I'm sick of the hype.

For the last 2 years, you've been sold a story: AI was going to replace 80% of developers. Code writes itself. Programmers are dead.

Then the data from the last 2 months dropped.

Builder.ai, a $1.5B startup backed by Microsoft, files for bankruptcy over fraud. A Veracode study reveals 45% of AI-generated code has security vulnerabilities. Another study by METR shows senior developers are 19% slower when using AI tools.

Does this mean AI is garbage? No.

It means the timeline is completely overblown.

We're breaking it all down. With data. No bullshit.

And when you're done, let me know in the poll below if you like this format. Your feedback matters.

Builder.ai promised the impossible.

Anyone, no coding skills required, could build a complete app. Just talk to "Natasha," their AI assistant, and she'd build everything. Mobile apps, websites, whatever you wanted.

Investors bought it.

Microsoft put in money. Qatar Investment Authority put in money. Total: $445 million in funding. Valuation: $1.5 billion.

The founder, Sachin Dev Duggal, was parading around Davos in January 2024 like the new AI messiah. Events with celebrities. BBC interviews. "AI is the cape that turns people into superheroes," he said.

Then the truth came out.

Natasha wasn't AI. It was a facade. Behind it were 700 engineers in India writing code by hand.

But the fraud went deeper.

Revenue inflated 4x. They projected $220M for 2024, reality was a fraction of that. Fake invoices with VerSe Innovation to pump the numbers. When investors started digging, everything collapsed.

February 2025: CEO out. May: bankruptcy.

Viola Credit seized $37M from their accounts. Amazon and Microsoft were claiming $85M and $30M in debts for cloud services.

The key point: Builder.ai didn't go under because "AI doesn't work." It went under because they sold smoke and committed fraud. But it's the perfect example of what happens when hype outpaces reality.

Now for the hard data.

Veracode published their GenAI Code Security Report in September 2025. They analyzed over 100 language models across 80 real coding tasks. Languages actually used in companies: Java, Python, C#, JavaScript.

The result: 45% of AI-generated code fails security tests.

These aren't minor bugs. These are vulnerabilities classified in the OWASP Top 10. The most critical ones that exist in web applications.

Why does this happen?

Because LLMs train on public code from GitHub and StackOverflow. And that code already has vulnerabilities. AI learns patterns, good and bad. Then replicates them.

It doesn't understand security. It doesn't understand context. It only predicts what code "makes sense" based on what it saw.

The important caveat: This doesn't mean AI is useless. It means you ALWAYS need to review what it generates. You can't copy-paste and call it a day. Anyone who tells you otherwise is lying.

This is the data that blew my mind.

METR (Model Evaluation & Threat Research) ran a study from February to June 2025. They recruited 16 experienced developers. People averaging 5 years working on large repositories (over 1 million lines of code).

They gave them 246 real tasks. Half with AI (Cursor Pro + Claude Sonnet), half without AI.

Developers predicted AI would make them 24% faster.

Reality: they were 19% slower.

But here's the kicker. After finishing, they still believed they'd been 20% faster. The productivity paradox.

Why does this happen?

Because generating code with a prompt feels fast. That dopamine hit of seeing code appear on screen. But then comes:

75% of developers said they read every line of AI code. 56% made major modifications. They accepted less than 44% of suggestions.

For juniors it works. Previous studies show 27-39% boosts for junior developers. AI teaches them, guides them, helps them learn.

For seniors, it's noise. They already know what they're doing. They already know their codebase. AI breaks their flow more than it helps.

Let's be clear.

AI is real. It's powerful. It's going to change everything. I'm not here to say it's garbage.

But we're not where they've sold you we are.

Here's the reality: AI speeds up certain tasks. The repetitive ones, the structured ones, the ones with clear patterns. That's where it works well.

But it doesn't replace human judgment. It doesn't understand complex context. It doesn't anticipate problems. It doesn't make strategic decisions. And it makes mistakes you have to fix later.

Who wins with this technology? The person who uses it smartly. Who knows when to apply it and when not to. Who reviews, verifies, and understands what they're doing.

Who loses? The person who thinks AI does everything on its own. Who delegates without supervision. The next Builder.ai.

We're in 2026, not 2030.

The timeline is inflated. The technology is real.

Don't confuse the two.

Did you like this different format? Let me know in the poll. Your feedback matters.