7 AI Startups Burned Through Over Half a Billion Dollars and Failed. This 23-Year-Old Spent $20 and Won.
770 people showed up to work at Builder.ai on a Tuesday morning in May.
By Tuesday afternoon, they were all unemployed.
Builder.ai, a company valued at $1.5 billion, backed by Microsoft and the government of Qatar, filed for bankruptcy. 770 people. Gone. One phone call.
And that’s just one of the seven AI startups that died in 2025.
Seven companies. Over $650 million in combined funding. Nearly 1,000 people total who lost their jobs. Some got zero notice. Others watched their companies implode over months of false hope and broken promises.
Meanwhile, KB is a post-grad architecture student who uses free AI tools and just 5x’d her workflow.
No funding. No pivot strategy. No board meetings.
Just a specific problem and some free software.
AI Concept of the Week: The $650 Million Problem Nobody Wanted Solved
Builder.ai ($450M raised, 770 jobs lost): Promised you could build apps “as easy as ordering pizza.” Most businesses don’t need custom apps. When Wall Street Journal investigated, they found the company’s “AI” was actually 700 human engineers in India manually coding everything. They kept that lie going for 8 years. Employees found out they were bankrupt during a company phone call. Many had been working with “zero dollars” in the bank accounts for days.
CodeParrot ($500K raised, YC-backed): Converted Figma designs into React code. Developers didn’t trust the code enough to use it in production. They pivoted so many times investors called it “pivot hell.” Made $1,500 in monthly revenue after 2.5 years before shutting down. Founder’s now on sabbatical after grinding for nothing.
Humane ($200M+ raised, multiple layoff rounds): The AI Pin. A $700 wearable device that was supposed to replace your phone. Ex-Apple team. Massive hype. Returns outpaced sales. The charging case was a fire hazard. HP bought some assets for $116M and immediately laid off everyone who worked on the actual product. Some employees got job offers at HP. Others got nothing.
Subtl.ai ($200K raised, 3-10 employees): Let you search internal documents with AI. Had clients like State Bank of India, but companies tried it and didn’t convert to paid plans. Accuracy issues. Too complex to integrate. Founder’s public quote: “I failed my team and investors.” Small team, but still people who lost their jobs.
Astra (backed by Perplexity’s founder): AI-powered sales automation. Launched in 2023. Dead by mid-2025. Co-founders couldn’t agree on growth pace. Enterprises didn’t want to give AI access to their Salesforce and Slack data. Lasted ONE YEAR. Team size unknown, but everyone lost their jobs over founder disagreements.
Wuri ($500K raised, YC-backed): Started as a no-code game creation platform for story-driven games. Then pivoted to transform web stories into audio-visual novels. High customer acquisition costs. Couldn’t scale. Founder’s quote: “Rapid change made it impossible to rely on past experience or traditional strategies.” They were chasing trends, not solving problems.
Locale.ai ($5M raised, 14-35 employees): Geospatial analytics for logistics and supply chains. Had international customers. Generated decent revenue. But after years of grinding through COVID and burnout, the founders returned money to investors and shut it down. They did it responsibly instead of dragging it out. Still, 14-35 people lost their jobs.
Now look at KB.
She’s recently graduated from a prestigious architecture school. She recently landed an amazing job in the town where she went to school. She builds 3D renders for actual construction projects. Her problem was specific: base renders from Enscape looked good, but they needed furniture, better lighting, and higher resolution to show clients.
Hiring someone to add all that? Expensive and slow.
She built a simple workflow:
Enscape creates the base 3D render
Gemini adds furniture and assets
ChatGPT upscales everything and makes it photorealistic
Total cost: Maybe $20/month.
Time saved: She 5x’d her output. What used to take her a full day now takes an hour.
Seven startups with over $650 million in combined funding failed in 2025.
One college student with $20/month succeeded.
Real Business Win: What Actually Makes AI Work
Dawid left a comment on my last newsletter: “The real adoption barrier isn’t between promise and reality. It’s between having a tool and having a reason to use it consistently.”
Builder.ai, CodeParrot, Humane, Wuri, and the others had tools. KB had a reason.
Those startups were solving problems that sounded good in pitch decks. They were chasing visions and trends.
KB was trying to finish her homework faster.
The companies that work aren’t asking “what can AI do?” They’re asking “what problem am I sick of dealing with?”
KB didn’t wake up thinking “I should use Gemini today.” She woke up thinking “I need to add furniture to 12 renders and I have a deadline.”
AI was just the fastest way to solve that specific problem.
Try This Today: Find Your KB Moment
Stop thinking about AI tools. Start thinking about your most annoying recurring task.
What task did you do last week that you’re going to have to do again next week? And the week after that? And every week for the next year?
That’s your target.
Maybe it’s:
Writing the same type of email 10 times a week
Reformatting data from one system into another
Creating social media posts from your blog content
Scheduling appointments back and forth over email
Pulling together weekly reports from different sources
Pick ONE. Not three. Not “my whole business.” One specific, annoying, repetitive task.
Now figure out if AI can do that specific thing.
Don’t build a custom solution. Don’t hire a consultant. Just see if ChatGPT, Claude, or Gemini can handle it.
KB didn’t need a $450 million platform. She needed three tools that already existed and a clear problem to solve.
Golden Nugget: The Human Cost of Hype
Builder.ai employees found out their company was bankrupt during a phone call. Many of them had been working for days with “zero dollars” in the company’s US and UK bank accounts. No warning. No severance for most. Just gone.
One engineer posted on LinkedIn: “Yesterday started as a normal day and by the end of it, we found out Builder.ai had filed for bankruptcy. Everything changed in an instant.”
These aren’t just numbers. These are people with mortgages. Kids in college. Rent due next week.
CodeParrot’s founders ground it out for 2.5 years. Pivoted. Tried everything. Made $1,500/month in revenue and had to shut down. Their team had to find new jobs in the middle of an AI hiring freeze.
Locale.ai’s founders were so burned out after years of nonstop work that they chose to shut down even though they saw new opportunities. They returned money to investors. Helped customers transition. But dozens of people still lost their jobs.
You know what’s working?
Boring invoice processing for accounting firms. Not sexy. Profitable. Nobody’s getting laid off.
Tools that help radiologists read scans 30% faster. Not revolutionary. Doctors pay for it. Stable jobs.
AI that books appointments for barbershops while the owner is mid-cut. Simple. Saves real money. The people who built it still have jobs.
The companies that survive aren’t the ones with the biggest vision. They’re the ones solving the most annoying problems.
And they’re not burning through hundreds of millions while nearly 1,000 people wait to see if they’ll have a job next month.
Builder.ai’s founder was at Davos in January 2024, talking about making people “superheroes.”
By May 2025, he’d laid off 770 people in a single phone call.
KB’s never been to Davos. She’s never raised funding. She’s never given a TED talk about the future of AI.
But she’s in the top 1% of people actually using AI for work.
She didn’t start with the technology. She started with the problem.
What’s the one task you’re doing this week that you’ve already done 100 times before?
Find that. Fix that. Forget the rest.
AI doesn’t fail. AI strategies fail when they’re built backwards.
And when AI strategies fail, real people lose their jobs. Nearly 1,000 in 2025 alone. And these are just the mainstream companies. That number’s going to keep climbing as long as companies keep chasing hype instead of solving problems.
See you next Wednesday, Dex.
The Chicago AI Guy
OH BTW - Research shows that 92% of AI startups fail. That’s higher than the regular tech startup failure rate of 63%. And 38% of AI failures? They launched products without market demand. They built first, then searched for customers. That’s backwards. And it cost nearly 1,000 people their jobs this year.