The Real AI Bottleneck Nobody Is Talking About

0:00Eric Siu:Does your experience with AI sound a little something like this? You've been prompting for 20 minutes. The output is polished, confident, and completely useless. That's what happens when AI doesn't know anything about your business. HubSpot's AI works with your actual customer data, every interaction and every signal your team has generated over time. So instead of prompting and hoping, you ask once and ship it. Check out HubSpot.com, the agentic customer platform for growing businesses. AI is cool and all that, but there's actually a bottleneck. Most companies are failing with AI if they don't really touch this.

0:31Eric Siu:So you can see, Neil, my good friend here, Gary Tan, said this is the actual bottleneck. The models are actually smart enough already. What is missing is a company-specific context locked in senior people's heads. So whoever cracks knowledge extraction at the company level unlocks the rest. As you work on this, please consider using Gbrain. So that's his product, right? Right. So but what really he's calling out here, what Gary Tan's calling out here is that Neil, for example, myself, we started with our organizations. You started since the beginning. I came a little later from my organization and I took it over.

1:06Eric Siu:But point is, you and I, we have the most context. But not only that, your CEO, for example, or other other executives at your company, they know the most and they can see more of the battlefield than most people. So how do you translate all that context into one area that that maybe some shared context that other people can access? because if they can pick that up, you're able to move a lot faster, right? So I think if you can extract knowledge at the company level and safely where there's appropriate permissioning and all that, I think you're going to be able to move a lot faster. And that's why, Neil, you see so many people right now building this whole company memory or company brain stuff.

1:42Eric Siu:I think everyone's trying to crack it right now and there's a lot of horizontal solutions. But anyway, I think it's something that most people should look at if they want to be able to move faster.

1:51Neil Patel:Yes. So that was his company. Did he end up raising money for his company through Y Combinator? No, he didn't because it's open source. Oh, cool. So he's just giving it away. He's giving it away. Correct. He's probably already rich from all the Y Combinator money. So is Paul Graham. People are like, ah, does Paul Graham have a lot of money? I'm like, it's not disclosed, but I bet you he has over a billion bucks.

2:10Eric Siu:Yeah. What is it? Airbnb, Dropbox. What else is there?

2:14Neil Patel:There's so many, dude. And there's so many companies that we haven't heard of that are in B2B that are just worth the killing and ruin Y Combinator.

2:21Eric Siu:Yeah. Anyway, it was a good model for them. But anyway, just make sure that you pay attention to that. If you don't, I think most companies will fall behind on that side. So, okay. I think, Neil, you'll like this piece because we've kind of talked about this a little bit. I'm curious because I'm in these AI chats as well. Now you see executives talking about how people are starting to rationalize their token cost, right? So if you look at this over here, so this is the founder of Lendi. So pull the trigger today and switch to 100 % of Lundi traffic to DeepSeek v4, churning from anthropic models, right?

2:53Eric Siu:So the cloud models. This saves us millions of dollars, and we're actually seeing an increase in performance on many core use cases. This has been transformative for the business. And then so Chamath here, he tweets this. He says, if chapter one of AI was all about broad open-ended experimentation, chapter two may well be about realism and rationalizing the cost of chapter one and what is sustainable moving forward. And I think you're seeing that with your clients. I'm seeing that with my clients as well, where they're just spending an arm and a leg. You see Microsoft like, oh, we got to cut. Now Uber's like, oh, we got to cut down, right?

3:25Eric Siu:It's not to say this stuff isn't productive, but you can't just spend and spend and spend. And so Sam Altman even said earlier this week, this is the first time they're seeing people tighten their belts.

3:35Neil Patel:Yes, and I think it's going to continue because these things are just so expensive. Anthropics signed a deal with SpaceX. I think it's like$15 billion a year, I believe it is,

3:45Eric Siu:for AI, I think they're paying them 1.52 billion a month, something like that. So yeah.

3:49Neil Patel:Okay. So call it around 19-ish billion a year or whatever. That's going to cause SpaceX stock to be even worth more once AI IPO. I'm still not buying it. But we got an email from Claude, you know, a little bit ago, or Anthropic, which is the parent company name. And our account manager there for our enterprise suite was helping us figure out how we can end up reducing costs and integrate them more. So I think they're already on top of it because they're seeing pressure from organizations. And we've been talking about this for a bit. This stuff is great, but there's a cost benefit analysis. If you can get 5x more done, but 5x more done increases revenue by 3%, but your costs go up another 15%, is it worth it or not?

4:40Neil Patel:And you have to make the calculations.

4:42Eric Siu:Yeah, so what I would say is this. I think the fact that they moved to DeepSeek, I think DeepSeek, you can pay for it or you can host it locally. I have Quinn on my computers over here. So there is a trade-off to all this. Obviously, my team can tell immediately when they get degraded to a lower level and then they start complaining. But I would also say that Cognition today, you should take a look at this, Neil. This just occurred to me right now. Cognition came out with a new way of measuring your token effectiveness, okay? So check this out. There's a new AI productivity guarantee, and I think we're going to see a lot more of this.

5:20Eric Siu:So I'm pulling this up right now. So Cognition says, AI should earn its keep. Introducing the AI productivity guarantee. If Devin, which is their product, delivers less engineering value than you're paying for, Cognition will fund your usage until it does up to$10 million. Okay. Wow. That's a good guarantee, man. That's a good offer. So it's time for the AI industry to stop maximizing tokens and start maximizing productive output, which is what we've been saying for a while. Like, it's not just about token maxing. It's about output maxing, right? So we built a system to estimate whether an agent's output was actually productive.

5:56Eric Siu:and if so, how long a human engineer would have taken to do the same work. Validated on real engineers' times estimates working at enterprise code bases. So they do a blog post on this, but I think you got to go in this direction if you want to stay alive because these other open source models are catching up too quickly.

6:12Neil Patel:Yeah, and I think these guys, I don't know which models, but companies are not only becoming more careful, we're hearing now from companies globally because we've all heard the US versus China thing. And a lot of US companies are saying, oh, yeah, you know, we want US backed or US built or whatnot. We got to be careful with the Chinese products. I'm just repeating what I'm hearing. We're now seeing a lot of companies when it comes to some of these AI models, whether it's for their engineers or whether it's for the marketers, a lot of them be like, oh, these Chinese models, these ones are a fraction of the cost.

6:50Neil Patel:Let's just try them out. Who cares if it was created by China or someone else, they're like, let's just save the money. And I believe whoever is going to have a good enough model that is cheap enough to help with marketers, in the end, a lot of it's going to come down to price. It's like, my team is on Microsoft. Why are we not on Google? I remember when we made the switch to Microsoft, so many people in the organization, this was years ago, was pinging me. They're like, Neil, we don't want to do this. Come on, can you convince some of the others? I stayed on G Suite because my email is neil at neilpatel.

7:22Neil Patel:My team has an NP digital email, so they all had a switch. I didn't switch, so I didn't have the problem. I let the company do what was best for the company financially. And you know what? The team got used to Outlook and they use it. And that happens for a lot of stuff. When I go speak at some of these Fortune 500 companies, I ask them, what LLM models do you use? A lot of them are in Microsoft's models. And I say, why? They're like, it's included. Same with people who are on Google's G Suite, like Gemini. it's included it's easier it's cheaper and companies are here to turn a profit especially publicly traded ones why wouldn't they want to save money yeah i mean my my take on all this

8:04Eric Siu:stuff is you know well i think it's it's cool to play with the open source models it's cool to play with the frontier models you have to figure out what works for you um and i think this output maxing thing is just going to become a lot bigger and uh if you can by the way if you're a services company and you're figuring out how because we talked to a company this week and the only thing they're focused on neil was was actually output maxing they're like dude we're spending so much on tokens now how can you help us optimize that so we said this i think a week or two ago like that that's going to become a service and by the way a lot of things that we never would have thought would become services are going to start to become services so that's why services is not going anywhere like i think agencies as long as you're on top of this stuff you're going to be good for a very long time quick break one of our click flow users just told me that they're saving 90 plus hours a month on content alone.

8:50Eric Siu:ClickFlow is our AI SEO product that actually does the work, not just here's what you should do. It does it. It creates content in your voice. It helps you with other areas such as finding the right internal links and creating FAQs. We spent years making it, perfecting it with our clients so it doesn't just sound like AI slop. You can get a 14-day free trial at clickflow.com. If you're serious about AI SEO, you should check it out. Back to the show. So maybe the last episode we can do here, Neil, is on this piece over here. Oh, this one's cool. So, Neil, do you want to hear about the podcast mentioned that drove$29 million in revenue?

9:25Eric Siu:Yes. Okay. So this is from Chamath. He got an email. So hi, JCal. Chamath is from the All In podcast. 55 days ago, you had a four-minute conversation about EAs and Athena. So Athena helps source executive assistance. Since then, we have closed$29 million in new revenue, seen a negative 53 % decrease. So a 53 % decrease in blended CAC, cost of acquiring a customer, and placed 683 Athena executive partners, 683 new jobs. And so I'm sure he probably has a stake in this company. Maybe, maybe not. But all that to say is whether you think about a mention on Joe Rogan, for example, or the All In podcast where you have a lot of executives listening to it, or you even think about what's the reason?

10:14Eric Siu:Oh, Jacob's partner, Abud, on Andrew Huberman's podcast. It's a matter of not just finding the reach, but finding the relevant audience. And then you should be able to crush it. Because now I think that peptide company, by the way, they're going to crush it.

10:27Neil Patel:So yeah, I saw the Sathena one and I was like, all right. It didn't say it explicitly, but I'm assuming they were mentioned on the All In podcast. And from there, it drove the revenue.

10:38Eric Siu:Correct. So I had listened to that episode and that's what prompted me to kind of sign up again and see how it goes. But yeah. And speaking of these executive assistants, so David Sachs, another member of the All In podcast. I'm going full All In right now, Neil. So David Sachs, Apollo's chief economist, zero evidence of AI-related job loss. What we talked about earlier. So I just want to pull up this data, and then we can move to the final topic here. So zero evidence of AI job-related losses. The chart below shows the weekly ADP employment data, and there's zero evidence of job losses because of AI.

11:12Eric Siu:Instead, many firms are hiring AI implementation experts and the data center buildout is building, putting upward pressure on salaries for AI experts and on prices of semiconductors, equipment and energy. The bottom line is that AI spending boom is stoking both employment and inflation. Wow. Okay. As a result, non-farm payrolls for May could come in significantly higher than$95 ,000 expected by the consensus. It is Jevin's paradox playing out in real time. Cheaper technology is creating more demand and more jobs. Look at this over here. this is the weekly employment starting from i guess last july leading up to now pretty crazy

11:46Neil Patel:the one thing i would say about the ai jobs you are seeing a lot of narratives about people losing their job because ai and companies like um atlassian lock uh which is jack dorsey's company saying that they're laying off people because of ai i don't believe it i believe a lot of these guys are laying people off because they're not getting into the growth numbers they want or their stock price went down, not because we're more AI efficient. And at the same time, from a marketing angle, imagine going to all your shareholders and using the marketing message of we're laying off employees because we're not growing as fast as we thought and we're not hitting numbers.

12:25Neil Patel:So we're going to go and cut costs. It's just going to hurt your numbers even more. It's easier to use the AI narrative and spin your marketing message than it is to just

12:35Eric Siu:say the truth. I would just also say one more thing around this. I think this proves that we talked about short-term job displacement. I think it's going to happen in some areas, but I think you're going to see more task losses versus job losses. So a lot of tasks that we had to do before, we're no longer going to have to do because we can just have the agents handling it. And I think, you know, again, it's task losses before job losses. I think that's going to be how it goes for the next couple of years. And final thing, Neil, because we've been talking about so much about AI, I feel like we should talk about some boring channels that still win in 2026.

13:10Eric Siu:So these are some boring channels in marketing that still win in 2026. I'll start with number one over here. Direct mail, meaning mail that you physical mail that you get. If I mail you, Neil, if I give you a UPS, if I FedEx something to you, you damn right, you're going to open that thing, right? It doesn't matter what it is. You're going to open it because it seems like it's important to you. So direct mail, what's old is now new again.

13:33Neil Patel:Yes. Number two, email and SMS. I'll combine them because some people just do SMS, some people just do email, some people do both. It works. Emails are still getting opened. They're still getting clicked. It's a very effective channel. I don't know why people wouldn't be leveraging email and SMS. They'll be like, oh, AI is going to just summarize the emails. We're still seeing it work. I saw an interesting stat published by High Level. Out of the emails they send, their emails get a click rate of roughly 6%, extremely high. You know, people wouldn't be clicking 6 % of the time if emails didn't work.

14:09Eric Siu:Yep. Last one we'll go with here before we take off here and let Neil go to bed. Live events. Dinners are not going anywhere. Conferences are not going anywhere. In fact, they're going to become more valuable. Relationships are not going anywhere as long as humans are around. And I think they're going to, it's going to be harder and harder to come up with relationships because a lot of the young people we talk to right now, the reason why they're going to college is because they know it's going to be harder and harder to form relationships and they want to form those relationships. So whether you can create dinners, happy hours, things like that, you can go to events, you can be that super connector, that's going to go a long way.

14:41Eric Siu:So that being said, go ahead, Neil.

14:42Neil Patel:Let's head number four. Last one. You guys all know this channel. It's SEO. People still value it by clicks. I'm telling you, if you look at your data and you do brand recall studies and you survey people on how they found you, even if you're getting less clicks, if you can rank in the organic listings and you can start ranking and being mentioned in the AI overviews. Even if you're not getting the clicks, people are seeing your brand. And if you go and look and ask them how they heard about you, you'll start seeing a lot of revenues coming from people seeing you on Google, but not actually clicking on over to your website

15:19Eric Siu:because you're being mentioned in the AI overviews. By the way, last full month, do you think you're getting net more leads or net less leads or the same?

15:27Neil Patel:net less leads, net higher quality leads substantially. Yes, yes.

15:32Eric Siu:That's so weird, but yes.

15:34Neil Patel:But it's not weird because what would happen before is someone would do a search. They see 10 blue links. They may click on four or five of them, fill out their information. Now people are going to traditional search. And you have to remember in traditional search, there's AI overviews, there's AI mode, or they're going to chat GPT. They're asking all their questions that LLMs are identifying which agencies that they should contact. So they're going and reaching out to less, but the leads are much more qualified.

16:03Eric Siu:Yes. I'm just saying it's weird for you and me because we come from the past, right? Where it's like, I'm just not used to it is all I have to say. Anyway, thank you all for listening. Hope you enjoyed it. Please don't forget to rate, view, and subscribe. And we will see you tomorrow.