23 AI Trends keeping me up at night

0:00What's up, everyone? Today I'm going to talk about all the things in AI that's keeping me up at night. I've got a giant list of just things that I can't stop thinking about, all the opportunities, some things that scare me, some ideas that you can take. And if you stick around to this whole episode, maybe you'll be an insomniac like me. Maybe it'll get those creative juices flowing. Maybe you'll have a better sense of why I'm so excited about where we're at right now and some things that also freak me out. so I just figured I'd go up go on here and sort of reflect on a bunch of the things that are that are keeping me up at night that are just making me motivated that are that are interesting to me and maybe it'll interest you too and you know if you're listening to this I think I think it probably will I think you know you're probably one of those people that you know see a lot of opportunity might be you know 90 % see a lot of opportunity 10 % a little scared but you're also looking for some ideas to help you move along and make progress.

1:09The first thing that's really, you know, I'm thinking a lot about right now is like the one hour company stack. So, you know, you grab an idea, you vibe code something, you build a landing page, you know, you add a stripe and you can get first customers. I mean, the fact that you can do this, just the fact that this exists, that you can go to ideabrowser.com, get a validated idea and just start vibe coding things with whatever vibe coding tool you want is mind blowing. The fact that you can create a company in a day. So I think that like from my perspective, I'm trying to think about how can I throw a lot more experiments against the wall?

1:49I don't want to just build one company, try it for six months. I want to just create a culture and a machine that I'm creating multiple companies, trying different things, same audience or multiple audiences. We're going to talk about audiences later. But this whole idea around the one-hour company stack, I can't stop thinking about it. Second thing is, if you think about the old timeline, and this relates to the first point, the old timeline of building a company was you had an idea, you hired devs, that took a few months. If you can find them, you built an MVP. If you're lucky, you can do that by month three.

2:26You launch it, maybe you go to product hunt. And then eventually you get to first revenue by month 12. In 2026, you can have an idea or grab an idea from Idea Browser by 9 a.m., have something built by 9.15 a.m., have a product built by 9.45, you get the first customer by 10, and you iterate by lunch. Someone is going to respond and be like, how is that possible? That's Vibe Coded Slop. There's a few reasons how this is possible. One is you're using a Vibe Coding Platform. I shouldn't even say a Vibe Coding Platform. You're using an agent engineering platform. You're using something like Cloud Code or there's competitors too.

3:13And you build something comprehensive. I mean, Codex has gotten pretty good. Google AI Studio has gotten pretty good. Cloud Code's gotten pretty good. So just the fact that you can do this with one of those tools is awesome. The second thing is, you do have to have an email list. You do have to have an audience. You do have to have some customers in order to actually get them. Because otherwise, as you know, finding the customers is really hard. But if you have been building distribution, And that's another thing that's been keeping me up at night is just using AI to build distribution. So yeah, the old time versus new timeline, definitely something I've been thinking about.

3:58The other thing that's keeping me up at night is this concept I'm calling ambient businesses. So ambient businesses that run with zero or very low daily human input. So having agents monitoring market, identifying opportunities, executing for you, handling customer service, and basically setting up a business that you just check in once every few days and just see what's going on. I think that we're going to get to a point where these ambient businesses or autonomous businesses are going to start doing seven, eight figures. So this whole concept is just really, really cool to me. I think we're really early.

4:43I think that a lot of the autonomous company builder softwares end up building very AI slop stuff. But I do think that this direction, I like to think about it as the arrow of progress. The arrow of progress is moving us in this direction of you're going to be building ambient or autonomous businesses. You won't need to check in every single hour. you're going to have checks and balances that are going to steer your agents in the right places. And I think there's just a huge opportunity there. This timeline, the agent economy timeline, is something that also keeps me up at night. So between 2009 and 2015, that really was the App Store era.

5:32People downloaded apps and humans operated them. Then in 2015 to 2024, I think the API economy really started to take over. Developers were wiring APIs together. I believe that 2025 to 2030, the agent economy is here. So agents discovering and hiring other agents on the fly. So fixed tech stacks dissolving. I actually think that there's a huge startup idea for someone to build the glass door of AI agents. So how do you create reputation around agents and who to hire? If someone could create a marketplace, sort of like a moltbook that was a social network or is a social network for agents that got acquired by Meta for allegedly$200 million.

6:24What is the glass door of AI agents? I know this sounds like such a far-fetched idea, but this is going to happen. I read a stat, I think it was by Gartner, 20 % of commerce by 2030 will be agent to agent, machine to machine. So how do you create startups that look at internet products and just create the agent version of it? So I think this market is going to 52 billion in 2030. Today there's 31 ,000 agent skills on marketplaces, but they're pretty bad. Most of them are actually garbage. So I think there's a huge opportunity to build skills, build agents, and it's giving me bags under my eyes.

7:17So this whole concept of agents hiring agents, CEO agents, sales agents, dev agents, marketing agents, I recently did a tutorial on how to use Paperclip, which speaks to this concept. It's an open source technology. Go check that out if you haven't. But basically creating an org chart, like a serverless function, where agents spin up subtasks and shut them down when they're done. This is just such an interesting idea. Instead of actually prompting via jobs to be done framework, it's like with humans, how do you actually just hire agents who manage other agents to get the job done. Super, super interesting and a lot of opportunity there.

8:02YC predicts that there's going to be 300 plus unicorns in vertical AI in this decade. So vertical software, huge business. Constellation Software, I think they own like 500 plus companies or something like that. they've built SaaS in the vertical software space. So super boring workflows in education, in defense, in all sorts of really boring stuff. And there's the same opportunity to build your own constellation software, but in vertical AI. So I think that if you're listening to this, thinking about what do you have an unfair advantage? What is your niche that you understand really well? Because I think that the people that are building in the vertical agent map, I think there's a lot of opportunity.

9:01The YCs of the world are going to focus on these big, big categories. Insurance, real estate, logistics, elder care, legal health care, sales. I'm not suggesting that you go and do that. I think that you should pick a wedge in a sub-niche of one of these vertical categories, start there and then sort of expand from there. There's going to be so much funding into these big categories. So you kind of want to just pick something that has a little less competition. But I do think that there's a lot of opportunity here. So how do you think about vertical SaaS versus vertical AI? This is something I'm thinking a lot about.

9:46Vertical SaaS captures a fraction of IT spend. You're generally selling software licenses. Humans are operating the tool and you're looking at the$10 to$100 million outcome usually. Of course, there's exceptions to that rule. Vertical AI taps directly into labor P &L. What I mean by that is you're building basically an agent as a software. Because you're doing that, the thing that companies are hiring you to do is what they would hire human beings to do. So the actual market for vertical AI is bigger than vertical SaaS. You're going to want to think about selling outcomes and results. Agents are doing the work.

10:30And because of this, I think that the outcomes in vertical AI are on average going to be bigger than they are in vertical SaaS. So unlike SaaS, which captures IT budget, vertical AI replaces headcount. And that's just a 10x bigger total addressable market. So yeah, if that's not keeping you up at night and all the opportunities here, I do not know what is. So what are some boring goldmine verticals? You want to look at stuff that just runs on phone calls and faxes and boring stuff. These are all big categories like insurance. but insurance still uses 30-year actuary tables, legal, logistics, elder care, government accounting, construction.

11:14You're going to want to pick stuff in these niches but sub-niched, very, very sub-niched down. And if I were you, I'd try to pick something that doesn't have a lot of red tape. Obviously selling to government is really hard. So things like that are probably less interesting But the more boring, the better, and the more niche, the better to start. So SaaS has really evolved in terms of how we've been pricing. So it used to be that you would do per seat licensing,$50 a user a month or whatever. All these big SaaS companies did this. And that's why you're seeing a huge correction in the stock market with, well, two reasons why you're seeing a huge correction in the stock market with SaaS companies.

12:03Some of these stocks are down 50%, 60%. Huge companies that were once trading at 12x revenue are now trading at 4x revenue on billions of dollars of revenue. And why is it happening? Well, it's happening because there's going to be less seats, one. And then two, people are scared that, or investors are scared that you can just vibe code these solutions. so yeah you know the evolution is going from went from seats to usage based so pay for what you consume to now you're just seeing a lot more outcome based stuff so pay per result delivered and why are you seeing that because you're getting you're having agents actually do the work so yeah Gartner says 40 % of enterprise SaaS shifts to outcome based by 2030 so seat base is going to decline from 21 to 15 % So where's the opportunity?

12:58Why is this keeping me up at night? What are ways to go and create outcome-based pay-per-per-result businesses right now? There's so many. And if you can be the first to market, you have an advantage there because when you're reaching out to people, be it cold, or even if you're just posting on your email newsletter or on your social media accounts, it's interesting to people, right? So you might be able to sell a lot more. So this whole seat-based versus outcome-based shift is just fascinating. The old way of$100 a seat a month, pay whether you use it or not. You pay 10 seats, you get$1 ,000 a month.

13:43And the value is unclear. I think we've all kind of felt this. There's some, I won't name names, but there's some software that my company, my holding company late checkout pays for. and I'm just like, are we getting the value out of this? We're paying thousands of dollars a month. So this whole idea around maybe you pay$1.50 per Resolve tickets, pay only for results and you have companies like Zendesk which is pretty big already doing this. 83 % of AI native SaaS has already switched. So my big point with all this is someone's going to build a billion dollar business doing nothing but converting legacy SaaS to outcome pricing.

14:29And I think there's just a lot of opportunity to help them do that. I also think there's a ton of opportunity to just go and build outcome-based startups on your own. Why even help them when you could just be incubating yourself? I definitely think there's something to be thinking about. So I have to mention, I do think that there's going to be somewhat of a SaaS graveyard, but I started to think about what is the framework for thinking about what's going to die. So I think generic CRMs are probably going to die. I'm not saying that Salesforce is going to die. I'm not saying that HubSpot is going to die.

15:12Those companies, you can see the writing on the wall and they're moving towards that future. but I am saying that if you're generic and you're not moving to this future you probably won't work. So agents are going to do it better. Basic analytics dashboard, those businesses probably not working because AI generates insights on demands. Template marketplaces, very tough business to be in because AI is generating custom templates instantly. Scheduling tools, I don't know. agents handle calendars natively and just basic customer support chatbots are already replacing this so what is going to survive here basically vertical workflow tools that pivot to agent companies infrastructure and data modes so there's basically this scarcity flip that's happening so what is being commoditized by AI I will code generic content, basic design, data entry, and routine analysis.

16:21So what is scarce and premium and what I keep thinking about, and a lot of people talk about this on Twitter, is that the value is going to migrate from execution to judgment. So creative judgment, human-made crafts, physical experiences. I'm looking at incubating stuff here. It's really a huge opportunity. Original weird thinking, just like being weird is going to sell in 2026 and beyond. Because a lot of these LLMs are just, they're not good at being weird. And you have a unique purview into life. You've experienced certain things. So like leaning into your weirdness and then proprietary data.

17:04What is the premium stack in terms of what are people going to value the most in an AI world is something that I'm thinking a lot about. And I think that the most premium is going to be human made. I don't know if you saw the Porsche campaign where they did a 100 % human made ad campaign. so it was like a race to establish an AI-free logo, I could see that luxury brands are going to lean into human-made and no AI involved. Think of certification labels like organic for food, right? No AI. So that's just something to think about in terms of ideas for building stuff. Premium, no AI involved, most premium.

17:55The premium would be AI-assisted but human-led. So having some human in the loop I think is going to be seen as premium in the AI age. So you get human taste with AI speed. And then I think a commodity from a perception perspective will be like, I'm buying just a fully AI service. And then I think there's sort of a race to zero pricing in some categories. I mentioned that I was interested in incubating and investing in basically IRL stuff. The reason why is when digital is infinite and AI-generated scarcity shifts to physical presence with other humans. So I think things like karaoke bars, escape rooms, immersive theater, co-working, live music, this whole experience economy is already here and accelerating.

18:53And there's just a ton of opportunity here. And that's keeping me up all night. Another interesting concept is this concept called founder-agent fit. When I was on the come up, so to speak, everyone just kept talking about, when I moved to Silicon Valley, everyone kept talking about founder market fit. Do you understand the customer and the market? Do you have some insight as you as the founder into the market? So if you were building a social network for college students, were you recently a college student or are you a college student? And I believe that where we're going is founder-agent fit.

19:39So can you orchestrate a fleet of agents towards your goal? So the bigger shift here is thinking as yourself as a film director. So a film director is not holding a camera. The film director is not acting. The film director is not writing the score. They get performances for actors. They're trying to get the most out of their actors. Now the actor is shifting from a person to a machine essentially. So I think that that founder skill, that founder agent fit is just an interesting shift that's happening. And if you're really good at building agents for a particular niche, managing them, getting the most out of them, then you have an unfair advantage.

20:32We talked a little about this with regards to paperclip and zero human companies, but this whole idea of a ghost team org chart, the fact that in the future I imagine a team page, you go to a website and you go to the about page and then you click the team section maybe and then you see all these people and their big smiles and you really get to know the team. But the future ghost team might be just a couple people and a bunch of AI agents, sales agents, content agents, customer support.

21:09Maybe you name them. Maybe they have personalities. Maybe you end up even creating images for who these people are. And eventually they're going to talk back to you. They're going to be able to video chat you. They're going to be able to send you voice notes. It's going to be in the ballpark of working with a human being. So this whole idea of ghost team org chart is crazy. As a person who's building a holding company and incubating businesses, I think there's going to be a lot more holding companies because you're going to own a bunch of AI native agent businesses in similar niches or the same niche.

21:53and you're going to have these ghost teams running it.

21:58So I remember Kevin Kelly talking about the 100 true, or the 1 ,000 true fans. And if you're listening to this, you probably have heard of that. But I think that it's not necessarily 1 ,000 true fans anymore in the AI age. And this is something I've been thinking about. I think it's more like the 100 true fans. Because agents are cutting your costs so dramatically that 100 people paying you is a real business. And because of the fact that you can build software and charge$1 ,000 a month or$500 a month because they are doing the work of potentially human beings, then there's a way to build a huge business there.

22:47Also, they don't even need to pay you that much. Because you can run it with agents and your team needs to be so small from a cost perspective, it could only be you. I just think that there's this world where you create these micro-monopoly math. I'll go through what I mean by this. Let's say you have a 5 ,000 engaged niche audience. You build a custom app. Maybe it's in 48 hours. If you have an audience or if you have a newsletter or something, you can get to 100 customers at$50 a month. But then you're running the business with agents. And then you're making about$60 ,000 of profit for one person, which is incredible.

23:32And then you can go and incubate multiple of these and expand this one. So yes, you do need to get 100 customers at$50 a month. And that's why I believe building media and building content and understanding how to build a machine that creates high quality meta ads is really helpful here. So even if you don't have an audience, you can pay for them. And yeah, it's going to cut into your profits, but so be it. I've been pretty optimistic up until now. But I will say one thing that kind of freaks me out is the agent attack surface. So I'm sure you've heard of prompt injections, things like poison context windows, malicious MCP service, agent-to-agent manipulation, permission escalation, compromised training data.

24:30Basically, because we're giving access to so much through our AI agents, I would be lying to you if I said this didn't freak me out, that bad things are going to happen. And I think bad things are going to happen. I think that cybersecurity hasn't caught up to how fast we're moving in this AI agent world. And because of that, I think some bad things are going to happen. So of course it's going to keep me up at night. Palo Alto Network's documented real world agent injection attacks. and if Palo Alto Networks is saying there's going to be a bunch of real world agent injection attacks well I definitely trust them.

25:19So how should we think about agent injection versus phishing? So in the past if you think about phishing like say 2010 it was basically like how do you trick a human being into clicking a bad link? targeting email inboxes, human judgment is the defense. So if you had a good eye for phishing, chances are you'd be okay. Even with that, billions were lost per year. So agent injection, where we are today as of recording this, you could trick an AI agent via hidden instructions. It targets context windows and web content. The agent autonomy is the vulnerability. and I believe that the potential is far bigger than phishing.

26:05So where agents have system access and make autonomous decisions, poisoning their context window, I guess, is the new phishing. I think it can be a lot more dangerous. I think a lot of bad things are going to happen. I do think there's a ton of opportunity to build cybersecurity software that helps with this. So that's a whole rabbit hole of startup ideas I can go down, but I think it's something to be aware of. The other thing to be thinking a little bit about is the agent permission stack. What can your agent access? Files, emails, calendars, bank accounts. People are giving bank account access to some of their agents.

26:48You're seeing them, here's$5 ,000, go and trade for me. What can your agent remember? Conversations, personal data, business data. What can your agent do? Sending emails, make purchases, modify code, delete data. What can your agent share, for example, with other agents or with third parties? The point here is you're going to want to do some digital hygiene. So quarterly agent cleanses, review permissions like you review app access on the web. Sometimes I'll go into some of the SaaS's I use and I'll be like, yeah, this particular app really doesn't need access to it. So I unlink it. And I think we're going to do a similar thing with agent permissions as well.

27:37I believe that now in the AJAI, the build cost is basically zero. Agents are doing a lot of the work. There's a ton of niches that are wide open and audiences are underpriced. I don't believe that this is going to last forever but this is what's motivating me so much. I think that there's 12 months where competition starts catching up. Some of the best niches get claimed. Some of the tools get crowded. I think there's probably 24 months where the window narrows. The builders who get started now are going to start owning moats around data, network, brand, trust. So people keep waiting for things to settle down.

28:18Things are not settling down. This is the new normal. I think that there's so much opportunity here and that's why every day matters so much. You definitely can listen to the podcast, Startup Ideas podcast, this podcast right here because I'm just sharing things in real time and just trying to help you increase your probability of success with ideas, tactics that I see working in real time. I believe that this window is asymmetric. So what I mean by that is what you need is an API key, some prompts, a tweet, a niche audience of, as we talked about today, pretty small, 100 to 5 ,000. And there's this asymmetry.

29:03So what you can get is a business that runs 24-7. You could create a business that has 95 % margins. It's not crazy if it's agent first. Okay, could it go to 70%, 80%, 60 %? but these are incredible businesses that you can be creating with compounding distribution and zero or a few employees. I believe that this is the most asymmetric time to be building a startup. What people are saying now is don't build in public, don't build in public. It used to be build in public, build in public, build in public. I think that it is helpful to build in public. People say don't build in public because you're inviting competition.

29:48I think that the benefits outweigh the cons.

29:54Specifically, if your followers and your audience is actually your customers, then you could share what you're building. The community could vote on what you're building. What's so cool about this AI age now is you can ship updates in a day or two days or five days. Users are basically becoming co-builders and that just increases trust and distribution compound. So it creates this really cool flywheel when you're building with your audience. I also believe that forking a business, like how you fork a repo on GitHub, is going to be very common. So in a world where you can just copy other people's businesses really quickly, bringing in a community and making them feel like they're a part of building what you're building I think is going to be absolutely a huge moat and really important.