Here we are, facing yet another con in the artificial intelligence market. This time the protagonist is a London startup called Builder.ai.
They’re one of those companies that offer something like Lovable: you chat with the assistant, and the assistant builds your applications. Vibe coding, pure vibing. Except it turns out that a large share of the problems this AI agent solves were not solved by artificial intelligence at all, but by 700 developers in India, paid offshore to solve problems passed off as the work of an AI.
A valuation made of confetti
This company was valued - in venture capital terms, its market value, which is all confetti anyway if you look closely at this blessed venture capital - at 1.5 billion, with more than 500 million invested. And they went around selling work done by poorly paid developers.
The funny thing is that this isn’t what got them in trouble. What got them in trouble is that, again to inflate the valuation - which is the essence of this AI bubble - they had set up an ancillary, parallel company and created a fake cycle of invoices to pump up the metrics of the money they were moving in the market. All to inflate the valuation, and it was all fake. That’s the reason the company was called into question. The thing about the offshore developers is secondary, nobody talks about it, but the real problem was the fake invoicing.
A story like this is interesting because it’s yet another example of what happens when you try to build a business, but that building gets bent by a capitalism that doesn’t look anyone in the face, where the valuation of an initiative matters more than what it actually does. It’s all baloney.
Not just startups
And it doesn’t only concern the startup world. Years ago, if you remember, there was Amazon Go, a service from Amazon, a much bigger company, a real empire. You walked into stores using this service, took things off the shelves, and there was no checkout at all. Cameras watched you take products from the shelves, the bill was computed automatically and charged to your credit card when you walked out. Spectacular.
Sure, but then it came out that in that case too there were annotators, again in India, watching the cameras and working out who took what from which shelf.
Why? As I imagine is also the case with Builder.ai: when an AI model isn’t able to do something, you can try to teach it. To do that, you first need to collect data. So in both of these cases of baloney, they went through deceptive data collection phases, with the well-meaning intent of eventually building that automation - but in the meantime they made some very big claims.
It would be nice to also talk about Mechanical Turk, Amazon’s annotation service, which played a role in the birth of deep learning: it was used to label ImageNet, the dataset behind AlexNet in 2012. There are many things I’d like to talk about, but since it’s Saturday I’ll leave you alone.
Substance, substance, substance
Please, folks, stay alert: baloney is always around the corner. Keep your head on your shoulders and don’t fall for the claims, especially from these ultra-capitalized startups, let alone what the giants say. Substance, substance, substance. Because artificial intelligence is real, but it has to be done with judgment.