YC founder thinks AI Wrappers are doomed to fail

[ams59]

Theo - T3 a YC founder says the following on his last video

AI, what's happening? AI is tiring. I am really excited for the things that AI technologies enable, but the conversation has gotten quite frustrating. The thing I've seen that has been the most painful is this idea of more and more AI companies.

I don't think AI companies are going to survive through 2024. AI is a powerful way for us to augment how people do things, and OpenAI and companies like it are going to do great. The providers of these technologies have a very happy path to success. What I don't see is tooling companies building entirely around AI for user products finding too much success over the next year. If every company is trying to make it easier to write code via AI, there's a feature like co-pilot that's going to squash them. For every company trying to do really good image generation with AI, Photoshop adds one feature and kills them.

I think the future isn't AI companies; it's AI features in existing company's products. At this point, the vast majority of us have used co-pilot, and I think co-pilot is still the golden example of AI done right. It takes something we're already doing, which is writing code in our editor, and uses AI to make us slightly more effective as we do that. That is really powerful. Even though I didn't think I would like it going in, I've been blown away with the experience I have had using co-pilot. It's made programming more fun and more efficient. Although some of the code it writes is jank, and I don't notice until it breaks something, it's a huge efficiency win. Most importantly, it doesn't get in my way or require me to change the tools I'm using entirely.

Again, if we compare this to other options in the industry, the expectation is this whole separate product or ecosystem will be moved to just to get these new AI benefits. I'll promise you something: I'm not changing the photo editing software I use just because another one adds AI to it or because some other AI-first solution exists. The winners aren't going to be the companies that create new AI tools from scratch. The winners are going to be the people who build AI into their existing products the best and most effectively. We're already seeing this with everyone from Adobe to Notion to GitHub and Microsoft meaningfully integrating AI into their products rather than selling AI as its own product.

I think that covers it. 2024 is looking really, really exciting. Clearly, I had a lot to say about it. This video is almost an hour of raw recording time, so sorry to my editor. Appreciate you a ton. Definitely keep an eye on the channel over this next year because we have some really exciting stuff coming.

What do you think I got wrong? Because there's a lot of hot takes in here, and I'm sure you guys are going to disagree with a number of them. So let me know in the comments what you agree with, what you disagree with, and what you think is going to change over the next year. Appreciate you guys a ton, as always. Check out the video where I talk all about the tech of 202 for here. See you in the next one. Peace, NS.

 

[sen_]

@ams59 Not a controversial opinion. Theo's right on this.

 

[frankmu]

@sen_ It was controversial during invites to the last batch on this sub. Anyone questioning YC's investment in hundreds of wrapper companies was downvoted to hell.

Now while I wouldn't wish failure on anyone, YC has undoubtedly made a mistake here. I'll be interested to see what kinds of pivots take place.

 

[sen_]

@frankmu well they surely made a big mistake to give our spots to glorified wrappers!! lol. not angry at all lol

 

[ams59]

@sen_ Totally

 

[mitchb]

@sen_ uncontroversial does not mean right. i would think he would be right if AI remains as something that could be splashed on in a product

 

[sen_]

@mitchb Sure. That's the comparator here... AI wrapper as just add salt had fastly diminishing value. But using AI in an innovative way to unlock new potential in tech will be a boon.

 

[shamakwa]

@ams59 I think people just need to think smaller and more niche. That's the key. Build for a community.

 

[loulou1]

@shamakwa This should be the most upvoted comment. I feel that most successful companies became that way by doing just this. If you can find something that is currently clunky and a pain in the ass and you can make it more efficient, fun, cost-effective/cost saving, then you have a winner.

I think more folks need to look into less glamorous niches as well. I work in Ag Tech and there's a ton of money getting thrown around in that space. Probably others like sewer/water and trash have the opportunity there as well.

 

[my2kidsmom]

@shamakwa Smaller niche like an app that exists in a platform marketplace is still viable but might not give desirable returns to a VC.

 

[shamakwa]

@my2kidsmom Sure, but who cares? In a smaller niche you retain more and earn more.

 

[outlawstate]

@shamakwa If you don't need capital, you don't care.

If you ever need any capital, you care. There is no real capital market for small (but supposedly lower risk) ideas.

 

[masako]

@shamakwa Yes, but those startups should not be raising money.

 

[jesusyeshuaislord]

@shamakwa Yes. A reason why SaaS platforms and their App Store “wrappers” (ie Salesforce AppExchange) can both be doing fine. Same analogy. And yes, the niche “wrappers” are the ones that stick.

 

[stagnate77]

@ams59 This assumes all these “AI feature” companies are looking to get unicorn status. They probably want an easy exit. Make a feature, turn it into a company, and sell for $1-10M exit!

I spoke to several YC founders who told me to my face they don’t care about a big valuation. They are just looking to leverage the YC brand for a quick easy exit.

 

[ams59]

He makes a solid argument

 

[andreic86]

@ams59 Truly said.

Being in the IT services for a quiet long time I am now seeing the changes for developing saas, paas etc applications and software development which must be integrated with ai tools for the benefit of the business growth.

But in the other hand they also showcase their products as ai products rather than core product. Which is one of the biggest drawbacks.

 

[iam815518]

@ams59 I am not an expert or anything, but I disagree with this point of view.

If you go back in history, you’ll see plenty of examples of people calling the term 'wrapper.’ A software can be a wrapper around a programming language, when databases came, all the companies using databases became wrappers around databases, saving and retrieving information. This is not a good way of going about the world. If you take away the hype created by AI labs, it’s also a technology like databases.

Before starting to build something using LLMs, we have to understand them. It’s not gonna take over the world, as some people claim. In fact, we now know large models are really good at memorizing, so they'll eventually be able to mimic reasoning someday. (It’s like that one kid in school who’s okay in math because they're really good at memorizing.) Obviously, some breakthrough can change that, but for now, it’s like your iPhone's autocomplete, but with more data than your text messages to suggest the next word. Don’t get me wrong, LLMs are a great innovation because before GPT-3, this type of system was only in our imagination.

Now, with all these tools, as a founder, you want to build something. Now you are back at the age-old software wars on how to build software that can be defendable and how can it solve something really well compared to any other existing product out there.

For example, Microsoft and Google both added AI to presentations, but if you look at a tool like Tome, it's more intuitive and easy to use. In my case, I’ve been using it since I found out about it. What they’ve done isn’t adding AI to power point, they’ve taken the old Microsoft PowerPoint and simplified it.

I think with anything one can argue why sometimes can’t be done or why something can be done and I’ve always been in the latter, maybe that’s why I see this problem in this way. Feel free to comment if you agree or not.

 

[edenspeak]

@iam815518 I actually hate this argument because it seems so bad faith. The term "wrapper" implies a certain level of depth, and I have to think that we all know this.

Otherwise we can go on forever talking about what's a wrapper of what. Spoken language is just a wrapper for phonemes and pointing at shit.

The reason AI companies get this level of skepticism is because there is often a lack of depth to them. They often ask their users to move their entire operations or system to their "platform" just to receive the benefit of what amounts to a single feature for their trouble.

I'm not going to move all my notes, syncing and plugins out of Obsidian and into some GhostwriterGPT app just for the benefit of being able to highlight my text and ask ChatGPT to give me some ideas about it or summarize what I just wrote to myself.

That's what people are getting at.

 

[iam815518]

@edenspeak I agree with some of your points, but I don't understand why this is in bad faith, to be honest. Regarding the Ghostwriter app, it seems like a bad UX decision from the founders. Let me elaborate on the wrapper part. Think of Evernote, the note-taking app that was so popular and was sold for a couple of billions. How do you think it started? Wasn't it basically taking the data you enter and storing it in a database? In today's terms, isn't it equivalent to calling OpenAI APIs? That's what I meant, you can't say something will fail just because it's a wrapper. If it's solving a problem for its users, I don't see a reason for it to fail. Plus most startups start with one small feature and they build over time.