Eric Siu and Neil Patel push back on headlines declaring AI is in decline, showing that while some startups like Cursor and Lovable dropped sharply, others like Anthropic keep accelerating. The real story: attention is shifting, not collapsing. They warn against chasing new AI tools and stress verifying outputs to avoid hallucinations in marketing work.
Chapters
I have a good topic for us to start with today. Are you ready?
Yes.
So this is the decline of AI to death of AI. So I'll slow down a little bit. So the decline of AI to death of AI. Cursor, which is at $500 million ARR right now, went from 62% growth to negative 18% growth in 6 months, okay? That's one. Number 2, Lovable, $200 million ARR. They just hit $200 million, right? They were like celebrating it on LinkedIn. Lovable, these are vibe coding apps, okay? So Cursor's more of IDE, but Lovable went from 207% growth To negative 50%. Okay, now senior engineers are— look, there's a chart over here which I'm going to show you in a second, but windsurf down from 104% growth. Remember they sold to Google, I think eventually they're trying to sell to OpenAI. Remember that? $4 billion or something. Negative 30%. Total category, 76% growth down to negative 15%. So senior engineers are like, see, this whole stuff is a bubble, it doesn't work, blah, blah, blah. But the bubble isn't actually popping because when you look a little closer, Base44 that was acquired by Wix, remember that one? There's a lot of these acquisitions, right? But they went from 950% growth to 1,161% growth in the same period. And then Claude Code exploded to $1 billion in revenue and it's not even on this chart, right? So this chart over here, you can see, I can't pull it up that closely, but all you need to see is that it's turning red. Like Lovable's turning red, the growth rates, right? You can see Cursor's turning red.
Their data's off.
So you look at all this over here and then people are just like, oh my God, it's dying, it's dying. But the reality is no, like I would say a lot of these people using these apps are pretty promiscuous. I'm pretty promiscuous. Like I'll jump over to Replit, I'll jump over to Lovable, I'll switch to whatever. I'll tell you about the latest tool, you'll tell me about the latest tool. So I think people are still experimenting a lot right now and I don't think these things are slowing down by any means.
I don't think they're slowing down, but whoever posted this, their data's off. 'Cause I know at least one of the numbers for one of the companies you mentioned, and I'm not gonna name which one, and the numbers you mentioned are totally different. than what their real numbers are. So whoever created this chart, they're using inaccurate data.
And he didn't call out— this guy Tobias didn't call out where he cited it from, but you have— this is 12-week change, okay? Lovable, Cursor, Bolt, whatever, Windsurf, all these, Cognition, all these things. And there's a lot of red in here, but when you— I think all you'd really need to look at is who's actually driving the most enterprise revenue right now. It's Anthropic. Right? Anthropic's enterprise coding, right? And so that revenue keeps going up and they've beaten everyone else right now. And you guys use it, we use it as well. So like, I think this is definitely not the decline of AI. It's just a little, you know, fear porn.
It is, but it's inaccurate fear data. And the reason I say that is I know the data for at least one of the companies in there. And I know for a matter of fact, they're growing. And their growth rate is really, really good still when you look at week over week or month over month, they're still in the green and this guy's saying they're in the red.
Yep. So, and then this brings up to another topic over here on your Wi-Fi is slow, so I can't even get to it, but let's go back to the AI really quickly.
People look at AI and they're going to end up saying, oh, these companies are slowing down. What they also don't take into account, how rare is it for a company like Lovable? I think you use a stat that they got to $200 million in ARR.
Yeah, $200 million in like 2 years or something like that.
Or I think it's less than that, right?
Yeah. And then $500 million for Cursor.
Yeah.
Yeah.
So How— watch, uh, Lovable. Uh, how long did it take Lovable to get to $100 million in ARR? I think it was in months. So they got to $100 million in, uh, 8 months. All right. Uh, $200 million, my guess is roughly a year or a little bit more. Uh, 4 to 5 months. So they went from $100 million in 8 months, another 4 to 5 months, let's just call it 4, to add another $100 million. So they went to $200 million in 1 year. Now, of course, you and I know that they spent time before that to actually build the product, but dude, it takes forever to market a company and get it to $100 million in revenue. And the problem that we're facing right now in the market is people have unrealistic growth expectations of businesses. $100 million or $200 million in annual recurring revenue in roughly 12 months is extremely, extremely, extremely rare. And now people are expecting marketers and entrepreneurs to figure out how to grow their companies at those speeds. So even if their growth slows down, it's still so fast compared to what normal companies are doing, especially ones that raise half a billion dollars or a billion dollars, even those, most of those companies aren't growing at that pace. It's okay. They're still growing fast and still somewhat successful. And the reason I say somewhat successful is most of these companies have a churn problem. I wouldn't say the Cursors of the world or that space because it's more sticky, but the Lovables of the world have much higher churn problems than the Cursors of the world, right? Just because of the industry that they're in.
You know, sometimes I listen to Jason Lemkin from Zaster, right? He's on with Harry Stebbings on his podcast. And, you know, the way VCs look at it now is like, oh, you got— you have to hit these crazy growth rates, right? Or else we're not going to look at you. And that's just not— sure, that's what— that's the VC game that they're playing right now. But you also have to live in, you know, what's the base reality for the majority of people, right? And so what we're seeing with Lovable or Mercor, the data labeling company, or Surge, which does a billion in revenue, In bootstrapped, by the way, in 4 years, right? It's a data labeling company. And so like we're seeing a lot of this happen right now. We're seeing a lot of this spend going towards this CapEx going towards infrastructure build and like chips and all that type of stuff. And so, you know, that's the gold rush that we're in right now, which is interesting because earlier this morning I was talking to someone from my team and he was just telling me he was getting a lot of job offers right now. Every day he's getting a lot of job offers, okay? And I'm like, why don't you just take one of them? He's like, because I love it here, right? Because you care about, you know, my family, blah, blah, blah, right? I'm like, oh, that's nice, right? What do we need to do to keep you forever? That's one of the questions I like asking. The market's hot right now, but don't expect it to— like, for the vast majority of people, you have to kind of ground yourself in reality. Just because you're seeing these headlines all the time in X doesn't mean that, like, how come we're not growing at like 1,000% or whatever, right? It's just, you know, this is the world that we're going into.
Yeah, it's unrealistic. And if you look at like those data labeling companies, it's A, a lot of manual work. They're offshoring to a lot of countries like India. And B, the margins are low on those kind of businesses. and see what they're doing it so the models can be better trained, but how long do they need to keep spending all that money on data labeling?
You know, actually, so I was listening to a pod with the Surge founder, right? And so the question was asked him like, how long do you keep, like, is this data labeling thing relevant for? He's like, as long as AI still needs to be trained before we hit AGI is how long, 'cause AGI is like you have an AI that can do everything, right? So he's like, until we reach that, which I think Ilya from one of the, or maybe Andrej Karpathy was like, it's gonna take us like 10, 15 years to hit AGI. So maybe 10, 15 years out.
Cigar butt business.
Yeah, yeah, cigar butt business, which is what Warren Buffett said, right? But again, Merckor, I think they're doing a couple hundred, if not like close to a billion in revenue surge. Again, bootstrapped to a billion. And you have Scale AI, which sold to Facebook or sold 49% to Facebook, right? Remember that one?
Pretty much sold is their way of skirting around regulations.
Yeah, yeah, yeah. All that to say is like, look, I, I don't think there's good— there's a decline in AI right now. I think you and I are very, very bullish about it. All right. So I wanted to take a moment to tell you about my podcast co-host Neil's agency called NP Digital, and they work with a whole host of global companies or global organization. Also, Neil has SEO tools such as Ubersuggest and AnswerThePublic. All you have to do is go to npdigital.com to learn more, and we'll see you on the other side. I think What's interesting to me too, I'm looking at data from Ramp. So Ramp is the credit card company, right?
Mm-hmm.
So this guy is always showing interesting data. So Ramp AI Index shows AI adoption held flat at 45% of businesses in November, driven by slight declines in the finance technology sectors. So if you look at the models, by the way, this is interesting. OpenAI actually declined by 1%, okay? Anthropic up 0.8%, Google up 0.7%. So if you look at it over here, this is the overall. This graph, okay, OpenAI, like you see the green ones kind of going down a little bit and then Anthropic is like boom, right? Which we just talked about. And then Google is like, oh, you see it's like, it's kind of like popping up.
So to clarify on this, this is Ramp, the credit card company, because they're seeing all the credit card spend. This is business adoption of AI and they're classifying it as people paying OpenAI like for ChatGPT subscriptions or the API or Anthropic or Google Gemini subscriptions, etc.
Yep.
Or Nano Banana or whatever you want to call it.
xAI is going up as well, like very slightly, but Google, like Nano Banana, like Gemini pops up, right? So like, look, it's like once something new comes up, boom, this pops up, this goes down over here, and then they need to launch something new. And then they pushed 5.2 out yesterday, right? But in Anthropic, right, they're just focused on enterprise. And so like there's like stories within this one. It's like if Anthropic, we're just going to focus on enterprise, Google, we're going to try to cover everything, maybe we'll give it away for free, and then our models are just going to kick butt. And then OpenAI, they just like, they need to catch up in terms of innovation, so.
Yeah, no, look, I agree with you on the adoption. I actually think AI adoption, we're seeing it slow down in marketing as well. Like I've talked to enough companies, you've talked to enough companies. We all know about the AI slop problem. Companies are working on fixing it. I don't know if you saw ChatGPT recently announced their new model through APIs. Google did the same day as well. And they're talking about how it hallucinates less. At least Google mentioned that. But this slop is a real problem. And what we're seeing is a shift in companies from a marketing aspect. If you look at early or late 2024 or parts of 2024 to early 2025 and even mid-2025, everyone was pushing hard as like, hey marketing, you need to be more efficient. You need to do more with less people. We expect better results, faster growth. Like, what are you doing with the AI? And now we're see— starting to see a shift within organizations of they're saying, hey, you need to use AI where it makes you more efficient and it can help you scale faster and save money. But stop trying out every new AI tool that just comes out because it's creating inefficiencies and we're not really seeing revenue growth. Go back to the fundamentals that are working and figure out how AI can make you more efficient in the fundamentals that are actually causing a lot of the revenue for the business.
Dude, when I was 23 years old, I was working for an agency called Wpromote, okay? So for this agency, I was an SEO link builder. Okay. That was my job. And I remember, because I was always pushing the team to try all these new tools, right? At a certain point, one of the guys messaged me, he emailed me. He's like, dude, like all you do is talk about new tools all the time, right? And like, that's great. But like, if we don't use these tools, like what's the point? And he was basically saying like, you know, stop doing this. Like, it's like you're mistaking motion for progress, right?
Mm-hmm.
And he was absolutely right. And don't get me wrong, I still love these tools, but like, this is what I kind of tell my team now. I'm like, hey, I get your testing all this stuff, but how does this scale over to our clients? For example, how does this help our customers? Because if you're doing it yourself in a vacuum, that's nice for you. And you know, you're showing some stuff off and you maybe think you're appeasing me because I talk about AI all the time, but you're not. It's got to help our customers.
And if you're going to have a similar approach to what Eric and I have, when Eric says help our customers, he's not talking about like, what are you doing for our customers? He's more focused on how is this impacting our customers' KPIs, right? Their key performance indicators. So someone in marketing will either want more rankings or more traffic or more conversions or more emails, whatever the number is. When we talk on this podcast about what is AI doing for our customers or ourselves, we really don't think about it from we got these images done or we got more content done or helped us with keyword research. We try to break it down to our KPIs and the KPIs are usually things that are tied to revenue growth or profitability.
Yep. By the way, so remember I told you about Granola, the little AI note-taker? So I have it on right now just for fun, right?
Okay.
Even though I'm supposed to ask you if we can record or tell you we're recording, but I got in an argument with a family member last week. I had it on, right? And I also did it for a leadership training thing. So let's do the leadership training thing. So there's business takeaways from this and then we'll use it for the personal side too. It's good for this. You can't use it with your spouse though, 'cause it'll never work 'cause you can't win those arguments. But that's for another day. So, we did a 3— we call it 3-year planning, but it's like an annual planning, okay? And I had Granola on, it was like an 8-hour meeting, it was on the whole time. And so, through that, it has all the transcripts from it. And then I basically, yes, you can summarize the answers and all that. But what I'm more curious about is like, hey, what is everyone's Achilles heels, right? Like, what is each person weak at? What is each— like, what's the one thing each person needs to address for 2026 to have a successful year, okay? And it was spot on. Like, for example, for me is like, Eric needs to stop opening the oven, right? Meaning that if you put a cake in it, you can't keep opening the oven all the time. You have to kind of like— You have to wait for it to cook. Yeah, yeah. So that's like, okay, I'm like, okay, my action is I'm not gonna add a new directive in Q1, okay? I'm also not going to help as much as I usually do and see how that goes, right? And then for someone else, it's like, oh, you need to be much tougher as a leader, otherwise everyone's gonna run you over, right? And so it's, then I got a nice little image for it and I just paste it into the leadership chat in Slack and now it's just sitting there. Nobody disagreed with it, right?
'Cause when you have a computer here, like you can't, what are you gonna do?
you at the computer, get mad at it.
Yeah.
And so it was very helpful across the board. It kind of rated each person in terms of like what their strengths are, what their weaknesses are. And it was very actionable too. And so, and then I could just keep asking all these questions. So that's on the business side.
Okay.
On the personal side, on the family side, I had it running and I was going on a walk with this person, right? And this individual would not take responsibility for the things that they've done and would like keep just blaming other people, right? And then at the very end of the conversation, this person's like, you know, I'm a very accountable person and I'm a very objective person, right? And then so then I just opened it up. I'm like, hey, how self-accountable is this person and how objective is this person? It was like 1 outta 10 on self-accountable and like 2 outta 10 on like objective, right? I was like, hey, what do you think about that? And this person can't get mad at me, right? And so he is like, okay, fine. Like I'll listen to it. Like, don't you think you'd be wrong sometimes? I'm like, well, just look at it, right? So anyway, all that to say is like helpful for conversations that you're having, helpful for leadership meetings that you're having.
So this Granola app, 'cause I haven't used it yet, how does it know that I'm talking versus you're talking? I get the, our voices are different.
Yeah.
But how does it know that you're Eric and I'm Neil?
I don't have a good answer for that, but I asked at the very end of the 8-hour meeting 'cause we had like 6 people in it, I think 5 or 6 people in it, right? Um, I was like, hey, who's speaker A, speaker B? 'Cause it'll label speaker A, speaker B, speaker C. It actually labeled everyone correctly 'cause I think it knows that like if, if I like so-and-so's saying, hey, what do you think about this? And then speaker C answers, then it starts to label that person. So got it.
So it'll identify a name as if I say Eric, it'll, and then you start talking, it'll know you're Eric.
And it's using like, I can use Opus 4.5, which is the newest Claude model. So that's what I use because it's a thinking model. So yeah, it knows.
Yeah.
Yeah. It's good. Like, because imagine all the Zoom meetings that you're doing and calls that you're doing and things like that. If you combine everything into one, you just have this brain that you can keep going back to. You already have pretty good memory, but what if you like have a second brain here?
On a side note, when we're working with charts and data, we find that Claude hallucinates less.
Claude is the best one right now.
Yeah.
Well, they say so. So OpenAI is like, they say they're the best. Everyone just keeps gaming the—
They all say they're the best.
So yeah, but we—
I'm not saying it is better or worse. I'm just speaking from a hallucination standpoint. When we actually give data, it comes back with the same data that we give it and it gives us the output that we're looking for. Okay. To be very clear, when we give ChatGPT the data, the output, A lot of times does not match the data we fed in.
Dude.
And this is not complex data.
You'll like this. So this might give you a little push to use MCP. So this week I got data from my sales team on like how much we're, you know, how much basically I was like working through the numbers on how much we can afford to pay for an SQL, an MQL, a lead, right? And then what I did was I took that data and then I took it from like HubSpot and Google Analytics data and I said, hey, based on all of our data, How much can we afford to pay for like an email, for example? And then it said, okay, you can actually afford to pay like $6 for an email if you want a 6-month payback. If you want a 3-month payback, you can afford to pay $3 per email. Well, guess what? Now I know specifically how much I can spend on marketing, right? If we can spend like, for round numbers, we can spend $100,000 on SQL, then we should be doing this. Like, it's once you know your numbers and this, you don't need to wait for a data analyst anymore and you can just keep messing with this all the time. It makes you a lot more confident with your marketing.
Did you double-check all the numbers when I gave you the output?
Yeah, it was, it was, this was using Opus 4.5. This is using Claude.
But it was accurate?
It was accurate, yeah.
Okay, cool.
Yeah.
Yeah, and that's what I hate. What I hate is when people on my team end up using these models and then they give me back the outputs and they're sharing it and they're doing, you know, a presentation to me or updating me on what they should do. And then I go and double-check. And I'm like, this is off. And at that point I'm really pissed.
So, you know something I'm working on now? I haven't done this yet, but I remember I read a book. This is like 21 or 22 years old, but I remember, I forgot what book this is, but I remember this boss would always make the people that were bringing work to him or presentations to him and always ask, say, hey Neil, is this the best work that you've done? Right? And I think today's version of that is like, hey, have you verified the data before you brought this to me? Have you double-checked this before you brought it to me? If not, then please go do it again and then bring it back to me when you're ready. And I don't think there's enough of that right now. I definitely don't do it.
Yeah, people know to bring me data that's checked. When they don't do it, I get pissed and they don't do it ever again.
But do heads roll?
If people keep doing it, yes, heads will roll.
So you've actually had to can a few people for this?
No, I haven't had to can anyone for it. But I would can it if they keep messing up and they don't check their data.
Yep, yep.
You give them the warnings and HR does whatever. That's the problem when you end up growing into an organization with a lot of people. And in general, we are small compared to the Walmarts of the world and stuff, but for most SMBs, we have a decent amount of headcount. You have to start following these—
Regulations, rules.
Bureaucracy. I wouldn't even call them regulations or rules. I would call them just internal bureaucracy and politics. And it's just like, okay, you go deal with this. I'm not going to deal with it.
That's where we're going to end because I really need to pee, so goodbye guys. Bye.
