See all Y Combinator transcripts on Youtube

youtube thumbnail

Critiquing AI Startup Websites with YC President Garry Tan

24 minutes

🇬🇧 English

S1

Speaker 1

00:00

Over the past few batches, we've seen a pretty big uptick in the number of AI startups that are applying to YC and building AI-focused companies. So today, we thought we would do an episode focused entirely on AI startup websites. Welcome to another episode of Design review. Today, we are very lucky to be joined by YC's president, Gary Tan.

S1

Speaker 1

00:26

Thanks for having me. So Gary, why did you want to focus on AI websites for this episode?

S2

Speaker 2

00:30

The number 1 question that is sort of the most important for a lot of startups is actually, why now? All of the activity around large language models, whether using open AI stuff or Anthropic or open source, it's just the most exciting why now that's happening because we just didn't have this level of capability. So when you have this crazy new why now, often you're actually literally trying to figure out how do you describe a new category that didn't exist before?

S2

Speaker 2

00:57

So that's some of the stuff I think we're gonna dive into today.

S1

Speaker 1

01:01

Awesome, so we're gonna get started with Rosebud.

S2

Speaker 2

01:03

All right, let's check it out. Right off the bat, what

S1

Speaker 1

01:05

do you see in here, Gary?

S2

Speaker 2

01:06

Well, there's a busy thing on the left that's sort of distracting me, but AI generated game assets to game development. Build games at the speed of thought, GPT to game. That's a lot of, that's 3 different concepts all at the same time.

S1

Speaker 1

01:23

If you had to take a guess at what they do based on this, what do you think it is?

S2

Speaker 2

01:26

I mean, I must be a game developer, so if I'm not a game developer, it's probably not for me.

S1

Speaker 1

01:32

Yeah. Yeah, I'm pulling out the AI-generated game assets and then game development. But I couldn't tell if game development is for game development or a play on words for, like, to game, like, development in general. And sometimes I think we see this from founders where they want to be cute with their tagline and sometimes the meaning gets lost Right, but they think it's clever because they understand what they do Right, but it makes it hard for new people to quickly understand

S2

Speaker 2

01:56

you don't want to be clever if clever equals confusing

S1

Speaker 1

02:00

And that's like the first thing that people are trying to understand when they hit a website is What is this and number 2 being is it for me? So I guess we're saying AI generated game assets for game developers, right build games at the speed of thought GPT to game so now it makes me think that like this helps you build games rather than just being for assets.

S2

Speaker 2

02:18

Yep. And then there's 2 obvious call to actions. 1 is a gaming asset generator, and then the other 1 is AI game dev platform. So it sounds like it is both.

S1

Speaker 1

02:28

OK.

S2

Speaker 2

02:29

Sounds like a lot of things.

S1

Speaker 1

02:31

Yeah. And so I'm confused. I don't know which to click. So maybe let's just keep scrolling and see if we can uncover this.

S1

Speaker 1

02:37

All right, so there's a bunch of media, so social proof kind of stuff. Pixel Vibe AI Game Asset Generator. OK, So now I'm starting to feel like maybe there's multiple products here. What do you think?

S2

Speaker 2

02:48

Yeah, for sure. And then sprites make sense.

S1

Speaker 1

02:50

And then there's more calls to action here.

S2

Speaker 2

02:53

Which is great. You know, tri-beta means, oh, this isn't vaporware.

S1

Speaker 1

02:56

So then I guess I click tri-beta.

S2

Speaker 2

02:58

Cool. Okay, this is Interesting. Describe your idea here. What would be helpful is like even a couple things to click on, like even a fly out here that's, you know, just give me some good ideas, just impress me, right?

S1

Speaker 1

03:11

Like how could you just have this right on the homepage so people don't even have to click? Every time you move people, make them do extra clicks, you have a risk of drop-off. Some percentage are not going to do it, and they're going to miss out on this.

S1

Speaker 1

03:23

Now we keep going, but wait, there's more. AI-generated skyboxes, talking heads, AI avatar animation. So it seems like there's at least 4 different products here, which is kind of confusing because usually you want a single call to action that you're driving people towards. Here, I don't know.

S1

Speaker 1

03:42

I guess there could be lots of different types of users coming here that may be interested in different things. But I don't know, what advice would you give to maybe improve what we're seeing here?

S2

Speaker 2

03:51

Seems like we have to be a little bit more opinionated about what we really want the user to click on and do. A good example of this is actually on this page. You know, if you really want people to try the beta, make that the call to action, and then make learn more instead of a button that's equal weight, just make it a link.

S2

Speaker 2

04:10

And 1 of the most important things to use is contrast in order to figure out, in order to be opinionated to the user. I want you to do this.

S1

Speaker 1

04:19

I would recommend for Rosebud to figure out what their main KPI is, and what outcome they want from people that hit their website. My hunch is it's not, go explore all of these products, or figure out whichever 1 is most interesting to you and try that. There's probably 1 that's their main priority here.

S1

Speaker 1

04:38

And given that this talks about game assets, my hunch is that's the 1, and it's the first 1 here. And so Maybe some of these other ones just get moved to secondary pages where it's not the main focus and you can keep everybody focused on just driving them towards the game assets, trying that out, getting them to use that if that is in fact the primary action that you want people to take.

S2

Speaker 2

04:59

The other question I'd have for them is, is there a way that you can make a video or Use animation that actually shows the product and then shows in particular for AI a feat of strength

S1

Speaker 1

05:12

Mm-hmm, so

S2

Speaker 2

05:12

it's like check this out. You've never been able to do this before You know type a few things and then look at this magic output. I mean, those are the types of demos that really wow people.

S2

Speaker 2

05:23

And that's actually why people are so obsessed with this technology now. Like, the demos and the technological feats of strength that people are capable of putting out, like that's just blowing people away.

S1

Speaker 1

05:34

I think the biggest piece of advice for Rosebud is to focus in on the primary product here that you want users to explore and spend time with and get them directly into that product to get the aha moment as quickly as possible.

S2

Speaker 2

05:49

That's great.

S1

Speaker 1

05:50

Yeah. Thank you, Rosebud. All right. Next up, we got MagicFlow.

S1

Speaker 1

05:54

So AI workflows made easy.

S2

Speaker 2

05:56

How do you feel about these call to actions right away? That's like create the free account.

S1

Speaker 1

06:02

I feel like the flow is not that people come in, they read this first paragraph, and they immediately click the button. I think what happens is they come in, they read this first paragraph, they're like, it is free, do I want to create a free account? And then they go down and explore more.

S2

Speaker 2

06:15

And then what are we looking at here? User image, you can remove the background. Okay, I mean this is actually pretty good.

S2

Speaker 2

06:22

It's sort of showing, hey this is what we can do. You identify, they identify the object, and then from there you can create different types of scenes. So this is interesting because I can't tell whether this is actually what the product does. Based on reading it, that isn't.

S2

Speaker 2

06:41

It's more trying to show an example of a workflow.

S1

Speaker 1

06:44

Which actually I think is important. Like, you know, we've worked with lots of kind of generic workflow companies in the past at YC, and I think the problem that they all run into is nobody's like, man, I really need workflow software. Instead, there's a specific problem that they're looking to solve.

S1

Speaker 1

07:00

Usually what works, to your point earlier, is to actually focus on very specific use cases that are the most common ones, and use that to get people into the product, and then they can say, oh, there's actually much more that I can do here too. So it seems like maybe the use case they're trying to target there is image customization. And maybe it does more than that,

S2

Speaker 2

07:20

I don't know. Okay, how it works. I really like when they do how it works very early because it brings you to a much higher understanding of what's going on.

S2

Speaker 2

07:29

It's kind of like when you do YC interviews and you get to do a demo. Drag and drop tool to create any AI workflow you can think of. And then I like that they're using Hugging Face and Stability AI and Chat. I mean, these are, you know, being able to pull models from Replicate, great.

S2

Speaker 2

07:45

Like most everyone's heard of Replicate at this point. And they know, OK, this is where I get a particular open source model to use.

S1

Speaker 1

07:51

And I think a lot of people will see those, and it will resonate like, oh, I'm using Chat GPT, or I'm using Hugging Face, or whatever. And that makes it feel like, oh, it's for me.

S2

Speaker 2

08:02

OK, iterate quickly with our step-by-step debug tool. That's actually fun.

S1

Speaker 1

08:07

All

S2

Speaker 2

08:07

right. OK, well, I'm skeptical of the claim about fun, but it sounds useful, so that's good. Workflows are so much fun. Optimize for production.

S2

Speaker 2

08:17

Make sure it's fast and cheap as possible. Well, this is great, but I think it is good to use very concrete claims, like it's 30% cheaper and 25% faster. You know, the tricky thing about this stuff is like, well, how do you know? And you know, there's not really a footnote.

S2

Speaker 2

08:31

I'm not saying you need to add 1, but those are sort of the questions that people sort of think through. And then you can use a simple API call from your app.

S1

Speaker 1

08:41

I think what I like about that section is it feels like it came from lots of conversations with users. Like, the faster, cheaper kind of thing is probably the thing that when they talk to users, they would tell them, were there concerns about using a product like theirs, or why don't I just build it myself, or something like that. So it seems like they've proactively addressed that, which is great.

S2

Speaker 2

09:02

How funny. I inherently tried to drag and drop because it felt like I should be able to do that.

S1

Speaker 1

09:08

They said that multiple times, drag and drop.

S2

Speaker 2

09:10

Literally, they said drag and drop. So the first thing I tried to do was, and then when I moused over this thing, it actually becomes a hand cursor, which is also the affordance for being able to drag. So

S1

Speaker 1

09:20

what happens when you click

S2

Speaker 2

09:21

on it? Absolutely nothing. And I actually clicked on it, and it dropped me into the login.

S2

Speaker 2

09:25

So I'm like, oh, I did the wrong thing. The problem with this interaction is it just doesn't build trust with me. Like, That's not what I expected.

S1

Speaker 1

09:32

Yeah, you just hit a wall and you have to turn back.

S2

Speaker 2

09:34

Well it said try, so okay. Okay, now you're just lying to me. Yeah, yeah, yeah.

S2

Speaker 2

09:42

This is interesting. I mean, I can see what they're trying to do. It's like trying to give, you know, you're getting tricked into the wall.

S1

Speaker 1

09:49

Nothing feels like a live example about this, right? Maybe something like Do your first 1 free like don't even actually force people to create an account and put that wall up front Instead you could just make it accessible And if somebody wants to save it or use it, or however they plan to use it, that's when you prompt them to actually create the account, because they're already invested in the product. When you click through, and it's like, try it now, and you hit this wall, there isn't even context about why you should sign up here that's encouraging you to actually like click the sign in button.

S2

Speaker 2

10:18

Yeah, basically you're always fighting the urge for the back button or to go look at Hacker News.

S1

Speaker 1

10:24

And how many times have we hit the back button already on this 1? Many times. My biggest piece of feedback for them is like there's great stuff here, It's just kind of small and like buried on the side there.

S1

Speaker 1

10:33

And the more that they can make that, like you want to hit people over the head with it. Otherwise they're just going to quickly scroll past it. If they can make it more obvious, then this seems great. Next up, we got Reality Defender.

S1

Speaker 1

10:44

Deep fake detection and protection. Stop deep fakes before they become a problem with Reality Defender's proactive detection platform. Alright, so what do we got going on in this video here? It's pulling out people and faces I guess.

S1

Speaker 1

11:00

Yeah, this makes me think this is like software for like closed circuit cameras. Right. Like to find people walking through the airport or something like that, but it talks about deep fakes. So it seems like maybe that's not actually what's happening here.

S2

Speaker 2

11:14

That's right. And there's really too much spacing between the lines of this header, and then there's not enough spacing between the logos and this bottom body area. So the interesting thing is up here, I'm sort of getting this enterprise salesy vibe, and then like it or not on the visual side, I think I subconsciously do this, and maybe this is a designer thing, but it's generally true that having really good spacing, it just sort of contributes to this feeling of quality.

S2

Speaker 2

11:45

And when you have little fit and finish issues like, you know, certain things being too close or headers having the wrong line height, it does ruin the illusion for me that this is something that I can trust at the enterprise level.

S1

Speaker 1

11:59

Yeah, it ruins some of the structure and organization.

S2

Speaker 2

12:02

Like there's way too much space between this and this. And you know, if the hard part about this is like, does it really matter? You know, the majority of people probably won't tell, but for some people It will matter and it just brings down the quality by that like little 5 or 10 percent.

S1

Speaker 1

12:20

Yep So I think we're like halfway down the page and we still don't really know what they're doing. Yeah, stop deep fakes. Okay Developed by okay.

S1

Speaker 1

12:28

So this is like a social proof like we're a legit team kind of thing. NATO uses it. Provides enterprises and entities of any size intuitive protection against damaging deepfakes. But I still don't know how or what they do.

S1

Speaker 1

12:42

Maybe they'll get into it here. Real-time results, okay, so it does something in real-time, we're still not sure exactly what it's doing in real-time though.

S2

Speaker 2

12:50

Yeah, we're halfway through the page and we're still really looking and we still don't quite understand what it is. And we're trying, and I think most users, when they hit this, they're not actually gonna try this hard. I mean, defending democracy sounds pretty important, but I also still don't know what it is.

S1

Speaker 1

13:04

How though?

S2

Speaker 2

13:04

How do

S1

Speaker 1

13:05

you defend democracy?

S2

Speaker 2

13:06

Authenticate, oh, so okay, there is an authentication piece. So we weren't totally wrong about facial detection, I guess.

S1

Speaker 1

13:14

Yeah. This is almost A123, maybe? Okay, so protect what's important. Scan user generated media for fabricated items Okay, so you scan user generated media you detect potentially damaging content before it spreads Verify voice prints detect doctor documents and grow your fraud prevention capabilities

S2

Speaker 2

13:32

Yeah, and so these this is interesting, but they don't quite go together. Because the first 2 sound like you're selling, it's an API of some sort, and you're selling to social media sites.

S1

Speaker 1

13:44

Scanner. Right. Or companies to detect fakes of their products, but then there's the people up top, so I'm still confused.

S2

Speaker 2

13:54

Yeah, I mean, that's 1 thing that we're always looking for in the homepage, like you want your ideal customer, your ICP to sort of look at it and say, oh, that's for me. And in order to have that, you actually have to write the home page for that particular customer, not like a nebulous, like any customer, anyone who might need our technology. Like being too general is actually a little bit problematic here.

S1

Speaker 1

14:18

Yeah, being specific is way better. And actually 1 of the best tactics to be able to do this, go talk to your existing users, ask them, how would you describe my product? And then take the exact phrase that they say and just put that up at the top of your site.

S1

Speaker 1

14:33

Because that's ideally the same thing that will resonate for other potential customers. What final words of wisdom do you have for a reality defender here?

S2

Speaker 2

14:41

I guess the main thing is, and it's always the main thing, which is, you know, this is my first conversation with this company and I really need to know who's it for, what does it do? And then again, like some of the things we said earlier still apply here. Like I kind of want to be impressed, especially when you're talking about AI, there's some sort of feat of strength that you can show.

S1

Speaker 1

15:01

It's like they skipped down to the lower tier things. Like the most important thing, who is it, who's it for? You gotta know that.

S1

Speaker 1

15:07

It's like Maslow's hierarchy, you can't progress to the next step. But they seem to do a good job at the social proof, you know, government approved, and it looks like government agencies are using this, and Visa, and Microsoft. So like, they sell me on that piece, which intrigues me to want to learn more, like I take it seriously, but I still don't know.

S2

Speaker 2

15:25

Yeah, I mean this case study might be pretty important.

S1

Speaker 1

15:28

Oh yeah, that's great. I totally missed that.

S2

Speaker 2

15:30

How a national public broadcaster detects deepfakes. Got it. I mean that sounds pretty important but even then there's not enough data to really make me want to put my info.

S1

Speaker 1

15:41

Yeah. Thank you Reality Defender. Alright next up we got Pump. Cheapest way to save 60% on AWS, the no-brainer, the fastest.

S2

Speaker 2

15:50

Okay, there is actually a video here. Let's check out the video. We are using Pump to save

S1

Speaker 1

15:56

60% on AWS. We are using Pump to save 60% on AWS.

S2

Speaker 2

16:04

I like the tunes. That's a pumped up to use this company. Exclamation point, group discounts.

S1

Speaker 1

16:16

AI, you know it's got to be good if it's got AI. I don't know. I think my takeaway from this video is a lot of sensory overload.

S2

Speaker 2

16:23

Yeah.

S1

Speaker 1

16:24

Like a lot of colors, a lot of movement, a lot of graphs. Like it's cool, but I'm not sure that I watch that and I'm like, oh yeah, I know what these guys do and I wanna use it.

S2

Speaker 2

16:35

Yeah, I don't know if I need to watch a Super Bowl ad for

S1

Speaker 1

16:38

it, right?

S2

Speaker 2

16:38

Like that's not the thing that's gonna get me to use it. I do like the call to action. View my savings makes it feel just very present to hand.

S2

Speaker 2

16:46

Like, oh, I could do it very quickly.

S1

Speaker 1

16:48

Yeah. And that seems like the main reason that somebody would use this, right, is to get savings on their cloud infrastructure. And so if you can, like that's the aha moment for this company. And so if they can show people right up front how much their savings would be, then it becomes a no-brainer to use this product probably.

S2

Speaker 2

17:05

I like the social proof. That makes sense. Gotta do that.

S1

Speaker 1

17:08

I think the visuals look nice. I think a lot of the bright colors and moving elements may be potentially overused. A lot of times animation can be used effectively to draw the user's eye to something.

S1

Speaker 1

17:22

And if a lot of things are moving, then it just becomes kind of overwhelming. And I think being selective about it. So it's, what's the thing that you really want to draw people's attention to, what's the next step you want them to take, and then put animation around that, drive a lot of the attention and focus to that. So we've seen a lot of these reduced cloud spend companies, and so I'm kind of wondering with my YC partner hat on, like what's different about what these guys are doing?

S1

Speaker 1

17:49

And probably that's something that potential customers will be thinking too. If you search for like cut cloud costs, probably dozens of potential products that you could go with to help with that. So like why these guys is probably the question a lot of their users are going to be asking.

S2

Speaker 2

18:02

So would you put like a X versus Y sort of reference to competitors?

S1

Speaker 1

18:07

Yeah, it's not clear to me what's different. Like I guess there's AI, but like lots of others are probably going to have AI too. So I'm not convinced that's enough of a differentiation.

S1

Speaker 1

18:15

They keep throwing out this Groupon thing, which makes me think there's some like group buying aspect where maybe it's like multiple companies pool, there, group discount, multiple companies pooling together in order to get on a better tier with lower pricing would be my hunch, but they don't actually really say that.

S2

Speaker 2

18:32

Yeah, that makes me want to click on FAQ. How does the group buying work? Yeah, OK.

S2

Speaker 2

18:37

So which is good. I mean, having an FAQ present to hand means that you can sort of pique someone's interest and then get their in. This is actually a good experience then. Like if within 2 clicks you can get, and actually having these 2 be the first 2, that's a good choice by them.

S1

Speaker 1

18:56

I think from the homepage, we were getting hints that there's this group discount, group buying thing that's going on here. We saw the group discount thing flash. We saw something about Groupon.

S1

Speaker 1

19:06

But they don't actually say that. And it feels like 1 of those things that you should beat people over the head with to make it super obvious that that's what's happening here and that's why it's better.

S2

Speaker 2

19:15

What would help me is if there was a customer that I had heard of that had used this already and saved X percent. Usually when that happens there's- There's

S1

Speaker 1

19:23

a bunch of logos right there.

S2

Speaker 2

19:25

Yeah, but there's nothing, you can't really, I mean, Ruby uses it, that sounds great. Yeah, I almost want there to be a case study with like whoever the number 1 customer is Okay, and then this is sort of their how it works view your savings estimate like the savings Let us know where you'd receive the dollars back Join your AWS account. Okay, there is a collective.

S1

Speaker 1

19:49

That seems to be the thing that's different here, and it's kind of buried. I think the advice on this 1 is to be more explicit around what's different here, how they actually help save 60%, reduce down some of the motion and animation and bright colors and instead use it sparingly to draw attention to specific things.

S2

Speaker 2

20:06

Yeah, less is more here.

S1

Speaker 1

20:08

Less is more. Yep, I think that's right. Okay, next up we have VoiceFlow.

S1

Speaker 1

20:12

And this 1 was actually submitted by 1 of our YouTube viewers, so thank you. Teams use VoiceFlow to design, test, and launch chat or voice assistants together faster at scale.

S2

Speaker 2

20:23

Start building. It's free. Or book a call.

S2

Speaker 2

20:28

This is pretty slick, I think.

S1

Speaker 1

20:31

This is a great demo. There's the start. So maybe that's...

S2

Speaker 2

20:34

That makes sense. Okay, I can visually understand that here's a flow chart. It's literally called voice flow.

S1

Speaker 1

20:43

It's moving a little too fast for me to dig in and be able to tell what is actually being connected, like what this flow is. And so that's a similar issue from the prior 1 that was about generic workflows, is trying to figure out, like, great, you do workflows for, I don't know if this is voice or chat, seems like maybe both. But to give a concrete example of who would use this and why, I think would be helpful.

S2

Speaker 2

21:05

I guess they're making a big deal of multiple people being on the same webpage, sort of collaboratively being able to build it, which is, that is a cool feature.

S1

Speaker 1

21:15

Okay, here's some social proof, I guess.

S2

Speaker 2

21:17

Just by 100,000 teams, that's nice. I actually like how chilled out this scroll is. This is a good contrast to the last 1.

S2

Speaker 2

21:24

It's like, boom, like so much movement. Like, oh, you know, I'm getting dizzy. And here it's like, oh, we're chill. Like, don't worry about it.

S1

Speaker 1

21:32

Yeah, it was the same thing up here. Like there's animation, there's motion, but it actually made us intrigued. It wasn't like, oh man, I need to put on my sunglasses.

S1

Speaker 1

21:41

That's right. And then down here, yeah, we can follow along. We can actually realize that there's a lot of companies that we've heard of before. Like that's impressive if all of these big companies are actually using their product.

S2

Speaker 2

21:51

Well, that's why they can be chill about it, Ben.

S1

Speaker 1

21:53

Maybe that's the difference.

S2

Speaker 2

21:55

Okay, the interaction image was so good, I sort of already got what it was for and whether or not it was for me.

S1

Speaker 1

22:02

That's how you know that it's well designed.

S2

Speaker 2

22:04

I'm getting the feeling that VoiceFlow existed for a while. So it is probably good if you're like 1 of the older AI companies that didn't just suddenly spawn into existence in the last 6 months, that, hey, we're sort of hip. We're with it.

S1

Speaker 1

22:18

Yeah. It seems nicely designed. It seems like lots of people have used it and given feedback on it, and they've iterated to this point, which is actually another good way to learn is to look at other products that have been around for a while and take some of the techniques that they use. And it's really easy to, like, they've already validated it.

S1

Speaker 1

22:35

They've spent years refining it. Like, you get to take their learnings for free right out of the gate. And then it's nice how they keep weaving the different logos and customers throughout as they're making a different point. And then they have somebody from that company that's giving a quote that reinforces that it's actually real people at this company.

S1

Speaker 1

22:54

It's not like 1 user tried it 1 time and then stopped using it afterwards.

S2

Speaker 2

22:58

Yeah, this makes it very, very trustworthy. And I like that this is a coherent point.

S1

Speaker 1

23:04

Mm-hmm. And they've got Molly's face or Colin's face there and it feels like real people

S2

Speaker 2

23:09

rather than something that's made up. Loved by designers. Great.

S2

Speaker 2

23:12

I mean, makes sense. We can tell. Mm-hmm. Yeah, game recognized game.

S2

Speaker 2

23:18

Well done, VoiceFlow.

S1

Speaker 1

23:19

Yeah, it was great. Clear call to actions, nicely designed, good hierarchy and flow. I think we understand what it is.

S1

Speaker 1

23:28

The right things drew our attention.

S2

Speaker 2

23:31

Yeah, I would totally use this if I needed a great AI assistant.

S1

Speaker 1

23:34

Yes. All right, that does it for this episode. Gary, thank you for joining us.

S2

Speaker 2

23:38

Thanks for having me.

S1

Speaker 1

23:39

And thanks to all the founders who submitted their websites for review. Just as a quick reminder, if you would like us to review your website in a future episode, go ahead and click on the link to the form in the description down below. So thank you all for watching and we'll see you on the next design review.