What VC's Are Looking for in AI Startups Today  🔗 **Source:** [What VC's Are Looking for in AI Startups Today](https://share.snipd.com/episode/8072959e-983f-48f1-b686-e4a19770aa03) ## Highlights - **Investors Want AI That Actually Finishes Work** - VCs favor AI that completes tasks, not chat overlays that only suggest ideas. - Jaeden Schafer contrasts a chat bar on a SaaS with AI that auto-generates titles, descriptions, and posts from audio files. Transcript: Jaeden Schafer And he says that the categories still getting the most interest are AI native infrastructure. So that's vertical SaaS built on proprietary data, systems of action that actually complete tasks and platforms embedded deeply into mission critical workflows. So basically, in other words, products that own something really essential. And there's a keyword I think he said in here that I 100% agree with. And that is, he said, AI that actually completes something. So I think there's a lot of this, a lot of these startups that were like, hey, look, we have like a SaaS, we have a tool, and then we stuck ChatGPT on top of it. And you could chat with ChatGPT, and it can give you like ideas about what you're looking at. In my opinion, that's very, I mean, basically, that's just the original SaaS. It's not super interesting. It might give you like some ideas or help you like troubleshoot or you don't need their customer support as much. But what I'm talking about when I, when I see AI and what I think a lot of these investors are looking for is AI that actually completes something in the past. Maybe I had to manually write a title and description for my podcast. And today AI can, you know, grab the transcripts of the audio file and do that for me. And if that's actually like accomplishing something for me, it's useful. Whereas if it was just like, I don't know, a chat bar on the side where it's like, you know, ask me what would be a good description for this type in your title, and I'll give you some ideas Like that is not useful. It's not automatically doing something for me. And that example I just gave you is like very basic. I mean, I think ideally, you would upload an audio file and it would fill out all the data and automatically find the best time to post it and look at your calendar and blah, blah, blah, Blah, blah. Like it's just going through and automatically doing stuff. ([Time 0:01:59](https://share.snipd.com/snip/037d728a-f68f-47b3-babd-6ba76515720b)) - **Avoid Thin Workflow Layers Without A Data Moat** - Avoid building thin workflow layers or generic horizontal tools without defensible data assets. - Abdul Abdirhan and Jaeden note investors want proprietary data moats like unique legal FAQ corpora or user behavior datasets. Transcript: Jaeden Schafer What are investors less interested in investing in AI right now? And what that is, is thin workflow layers. So generic horizontal tools, light product management software, and surface level analytics. If an AI agent can replicate the core value quickly, I think investors are not seeing this as very defensible. So maybe even some of my previous examples weren't the greatest because in a sense, they want your what they what they want tools to be able to do is have some sort of custom data set some Sort of um you know deep integration into something that's super super critical and it's not something that just like a chat gpt or anthropic can can replicate easily and the reason Being anthropic right now is rolling out all of these new these new tools right they're doing like anthropropoc for Finance and Anthropoc for Legal. And basically, you have a company like even Harvey AI, who's rated a ton of money. And I'm hearing, you know, anecdotal stories from people who are saying, like, you know, I've used Harvey for my law firm. And now I'm using Anthropoc, just rolled out Anthropoc for Legal. And I don't need this whole other tool. I just use my Anthropoc account. And all of a sudden, this is just as good as Harvey. So it's really interesting what is actually being seen as defensible today. So Abdul Abdirhan of F Prime added that vertical software without any proprietary data motes is no longer super compelling. So they actually want you to have a data moat. Maybe you ran a, you know, an FAQ legal website. So you have all of this data on, you know, legal FAQ questions. I've seen some startups do this. And that is like a proprietary data moat where maybe you have, you know, information that no one else has, or maybe you have information because you have a company and you can see what Your users, you know, what their behavior is. And so that could be a proprietary data mode. ([Time 0:03:40](https://share.snipd.com/snip/d3af16ae-ddcb-43bd-b4d4-c547eb371da0)) - **Shallow Product Depth Is A Red Flag** - Surface-level UI improvements and automation are weak differentiation because barriers to entry dropped. - Igor Ryabensky warns shallow product depth is a red flag and true workflow ownership matters from day one. Transcript: Jaeden Schafer Igor Ryabensky of Altal R Capital said that basically was arguing that shallow product depth is a really red flag, a big red flag. He said, if your differentiation mostly lives in UI and automation, that's no longer enough. The barrier to entry has dropped, which makes building a real moat a lot harder. I think for new companies, that means that building around, you know, true workflow ownership and a really clear understanding of the problem from day one is super, super important. Massive code bases are not an advantage anymore, right? Just being like, look, we have this massive code base we've worked on for a long time. Speed, focus, adaptability, all of those things I think I would argue are much more important. ([Time 0:05:33](https://share.snipd.com/snip/72e1c283-d25f-461f-ab99-76679cffcbe9)) - **Execution Beats Process For Developer Tools** - Developers increasingly choose execution-focused tools that finish work rather than mere workflow managers. - Jaeden cites Cursor versus Cloud Code as an example where one owns the workflow, the other executes the task. Transcript: Jaeden Schafer And he's like just showing you what it looks like through the lens of developer tools. So he pointed to the contrast between Cursor and Cloud Code as kind of where things are going. He said one owns the developer's workflow, the other just executes the task, which I think basically shows that increasingly developers are choosing execution over process, right? Like they don't just want you to own the workflow, they want you to actually get the thing done, which is what CloudCode is doing really, really well right now. It's the number one tool we use at AIbox. ([Time 0:06:56](https://share.snipd.com/snip/a8cef705-bb90-4a06-849e-1fde04fca2e1)) - **Agents Erode Interface-Based Moats** - If AI agents can perform tasks directly, human-interface-driven workflow stickiness weakens. - Jaeden explains agents controlling screens or using model context protocols could replace Zapier-style integrations. Transcript: Jaeden Schafer If agents are doing the work, then a lot of the traditional kind of workflow stickiness is becoming a lot less relevant. In the past, getting humans to operate inside of your software was a pretty powerful moat. Now, if an agent can perform the task directly, then owning the human interface doesn't actually matter that much, in my opinion. I think you're looking at a lot of these integrations that, you know, you pay companies like Zapier and Bubble and Make. I think a lot of those are losing their edge, right? Because as you have things like model context protocol, it's going to make it a lot easier for the AI agents to just go in and link directly to the software. And so instead of me having to go and set up some sort of integration straight into my, you know, meta ads account that, you know, it's kind of create this automation and I got to go tweak It and it's really complex. Instead, I could just go to my agent that's running on my computer, my like OpenClaw or whatever, and say like, hey, go to my meta account. This is the login. Go change these things. Go make these tweaks. And you don't need this kind of like integration. You don't need this Zapier because the agent is just taking over your control of your screen and the integration is just literally the computer being taken over. ([Time 0:07:27](https://share.snipd.com/snip/4bc5f53c-da20-4452-9ec2-40911364b3cd)) - **Thin Wrappers Can Win With Viral Growth** - Calorie-tracking app Calory (Cal AI) succeeded as a thin wrapper by dominating TikTok growth and reached $30M ARR and 15M downloads. - Jaeden uses this exit to show thin wrappers can still be acquired if they achieve scale. Transcript: Jaeden Schafer And Cal AI just helps you track your calories. It helps you lose weight. And a lot of people are like, oh, man, this is just a thin, you know, wrapper on top of a chat GPT. But guess what? They had 15 million downloads. They had over 30 million dollars in annualized revenue. And really, a lot of their unlock was that they really hacked the growth hacking on social media, TikTok and making shorts. And so I think on the one hand, you have a lot of these people that are saying, you know, oh, look, like you can't invest in these companies that are thin kind of wrappers. Well, if you have another angle, like if you have the growth kind of locked in and you can get $30 million annual recurring revenue, guess what? You're going to get an acquisition. And so that's what happened. MyFitnessPal went and acquired them. So I do think that there is some interesting points here. And you know, did they raise an insane amount of VC funding? I mean, not necessarily. That's not something that you have to do if you can scale it without it. ([Time 0:09:22](https://share.snipd.com/snip/8001ab13-d93d-424d-8e5c-58ba6e2b53ee)) - **Build Depth Through Data Ownership And Domain Expertise** - Build depth: own workflows, control data, and embed domain expertise to attract VC funding. - Jaeden stresses investors are reallocating capital toward companies with proprietary data and real domain assets. Transcript: Jaeden Schafer It's kind of the ownership of workflows. It's kind of the control of data. It's a lot of real domain expertise that they'll pay for. If you're an expert in a specific area, I think investors are reallocating capital towards businesses that have a lot of those assets and they're taking it away from products that can Be copied with very minimal effort. I think in a world that is, you know, quickly becoming AI first, becoming different isn't just about kind of adding an automation. It's about really owning something that agents can't replace easily. ([Time 0:10:23](https://share.snipd.com/snip/d8690f62-08e3-4718-9872-40fd99571283))