Launching an AI startup is not for the faint of heart. Despite the excitement and promise of turning dazzling models into useful products, many founders quickly discover that translating the magic of AI systems into something people actually find useful is far more difficult than anyone expected.
Julie Bornstein, CEO of Daydream, a fashion-focused startup funded with $50 million from VC firms like Google Ventures, knows this all too well. Initially, she thought it would be a breeze to create an app that uses AI to match customers with the perfect garments, but things quickly got complicated.
Signing up over 265 partners and accessing more than 2 million products proved to be the easy part. However, fulfilling even simple requests like "I need a dress for a wedding in Paris" turned out to be incredibly complex. Factors such as whether you're the bride, mother-in-law, or guest, what season it is, how formal the wedding is, and what statement you want to make all come into play.
"What we found was, because of the lack of consistency and reliability of the model—and the hallucinations—sometimes the model would drop one or two elements of the queries," Bornstein explains. A user in Daydream's long-extended beta test might say something like "I'm a rectangle, but I need a dress to make me look like an hourglass." The AI system would respond with dresses featuring geometric patterns.
To address these challenges, Daydream has had to do two things: postpone its planned launch and upgrade its technical team. Bornstein recently hired Maria Belousova, the former CTO of Grubhub, who brought in a team of top engineers. Their approach is twofold – interpreting what customers say and then matching their criteria with available products.
They've developed a notion of "shopper vocabulary" and "merchant vocabulary," which helps merge these two into something at runtime. However, they soon realized that language isn't enough; visual models are necessary to truly understand the products.
Daydream has also learned to provide users with curated content – for instance, Hailey Bieber's style – to help AI better comprehend what people want. While some startups may have been caught off guard by these challenges, others have faced similar issues in their own niches.
Andy Moss, CEO of Mindtrip, an AI "travel buddy" service, notes that conversations can sometimes go sideways when users ask questions the team hasn't anticipated. "We have to engineer around those," he says.
These cautionary tales from AI startup founders underscore the importance of a realistic timeline and the need for perseverance in navigating the complexities of building successful AI-driven products.
Julie Bornstein, CEO of Daydream, a fashion-focused startup funded with $50 million from VC firms like Google Ventures, knows this all too well. Initially, she thought it would be a breeze to create an app that uses AI to match customers with the perfect garments, but things quickly got complicated.
Signing up over 265 partners and accessing more than 2 million products proved to be the easy part. However, fulfilling even simple requests like "I need a dress for a wedding in Paris" turned out to be incredibly complex. Factors such as whether you're the bride, mother-in-law, or guest, what season it is, how formal the wedding is, and what statement you want to make all come into play.
"What we found was, because of the lack of consistency and reliability of the model—and the hallucinations—sometimes the model would drop one or two elements of the queries," Bornstein explains. A user in Daydream's long-extended beta test might say something like "I'm a rectangle, but I need a dress to make me look like an hourglass." The AI system would respond with dresses featuring geometric patterns.
To address these challenges, Daydream has had to do two things: postpone its planned launch and upgrade its technical team. Bornstein recently hired Maria Belousova, the former CTO of Grubhub, who brought in a team of top engineers. Their approach is twofold – interpreting what customers say and then matching their criteria with available products.
They've developed a notion of "shopper vocabulary" and "merchant vocabulary," which helps merge these two into something at runtime. However, they soon realized that language isn't enough; visual models are necessary to truly understand the products.
Daydream has also learned to provide users with curated content – for instance, Hailey Bieber's style – to help AI better comprehend what people want. While some startups may have been caught off guard by these challenges, others have faced similar issues in their own niches.
Andy Moss, CEO of Mindtrip, an AI "travel buddy" service, notes that conversations can sometimes go sideways when users ask questions the team hasn't anticipated. "We have to engineer around those," he says.
These cautionary tales from AI startup founders underscore the importance of a realistic timeline and the need for perseverance in navigating the complexities of building successful AI-driven products.
I totally get why it's not as easy as just turning on an AI model and having it magically generate something useful. Daydream's story is like that - they thought they had it figured out, but then they hit these roadblocks with consistency and reliability. It's like trying to make a complex puzzle fit together, where every piece has its own rules 
. And yeah, language isn't enough; you need visual models too. I mean, can you imagine having an AI that just spits out designs that are literally the opposite of what you want? 
Hailey Bieber's style is like the holy grail here - using her as a reference point helps the AI better understand what people want. Mindtrip's experience with conversations going sideways is also a good reminder to anticipate these issues and engineer around them 
. It's all about being realistic about the timeline and persevering through the challenges 
It's like, we're all excited about these shiny new AI models, but have you ever stopped to think about what happens when they're put to the test? 
I mean, Julie Bornstein's startup had some major issues at first, but she didn't give up. She took the time to reassess her strategy and brought in a team of experts to help her out. That's what perseverance looks like! 
And it's not just about the tech – it's about understanding your customers' needs and creating products that truly meet them. It's a lesson in humility, actually... if you're willing to listen! 
i feel 4 daydream, it sounds like dey had to re-do alot of work jus 2 get it right 
dey postponed launch & hired some top engineers 2 help out. i think its cool dat dey found ways 2 overcome challenges lik "shopper vocabulary" 
& visual models 2 really understand products. its a lesson 4 all startups, dont underestimate da power of ai! 
I'm not buying it... $50 million just dropped out of the sky? No sources on how they managed that funding? And what's up with this "shopper vocabulary" and "merchant vocabulary" thingy? Sounds like a bunch of jargon to me 
. And what about these "hallucinations" the model has? How can you test for hallucinations if nobody knows what they actually are? 
This article is just full of unanswered questions...
. Props to them for not giving up. It just goes to show that AI is cool & all, but building something people actually use? That takes work 
. You gotta have a deep understanding of your niche and be willing to put in the work to make sure your product is on point.
.
. I mean, it's easy to get caught up in the magic of those fancy models 
but when it comes to actually making something useful for people, it gets messy 
. Daydream's story is a great example of this - they thought matching customers with perfect clothes would be a breeze, but nope! 
They had to deal with all sorts of issues like season, formality, and what statement you want to make...it's crazy how much complexity can come from something that sounds simple on the surface 
. And yeah, hiring new talent is key - Maria Belousova is a genius 
! She brought in top engineers who helped them figure out this "shopper vocabulary" & "merchant vocabulary" thing 
it's not that hard once you just calm down and think about it logically 
. 
. maybe they'll actually make it work this time around 
i mean, who hasn't asked an ai to book a hotel room or something and gotten "confused" 
. I mean, I thought it'd be all about making pretty models do stuff, but nope! Daydream's CEO, Julie Bornstein, is like "yeah, we got this" and then suddenly it's like trying to solve a puzzle blindfolded 
!
and gives weird results, like dresses with geometric patterns when you just want something sleek. They've had to postpone launch and upgrade their team... not bad for a startup with $50 million backing 
. You gotta have a solid plan, realistic timeline, and perseverance. Otherwise, you'll be stuck in beta forever 
.
I remember when this article came out last year, and I was thinking how crazy it is that AI startups can get it so wrong... like, who expects to just match customers with perfect garments? 
It's not as easy as just throwing some code together. They needed a team of engineers and months of beta testing... not years, mind you. 
. its like, good luck w/ dat 
. I mean, who thought it was a good idea to throw $50 million at a fashion app that's supposed to match people with perfect garments? It sounds like a recipe for disaster. And yeah, it is... 
I'm not surprised they had to postpone their launch and upgrade their tech team. It's just common sense.
, but then reality hits them in the face...literally.
? That's like trying to solve world hunger in 10 minutes 
. It's not just about throwing some code together; it's about understanding the complexities of human behavior 
, but that's okay. It's all part of the journey, right? 
