Misleading title. Has nothing to say about working, i.e paid employment, with AI apps.
The main claim in the post: Their portfolio companies have shown an improved rate of accumulating revenue ever since LLMs took off.
Weakest part of the post: No attempt at explaining how or why a LLM affects these numbers. They allude to 'shipping speed' and 'product iteration', but how an LLM helps these functions is left unexplored.
There's an implied deductive argument that a LLM can write some code, so obviously shipping speed is faster, so obviously revenue is faster. But the argument is never explored for magnitude of effect or defended against examples where shipping faster or using LLMs doesn't equal faster revenue.
Also, nothing about sampling bias, size or spread.
Overall: Probably meant as a confidence boost to the sleep-deprived founders out there. But teaches nothing.
Reminds me so much of the no-code bubble maybe 10 years ago, which also was full of big loud talk about shipping speed, product iteration, and developer obsolescence.
Exactly this. They say increased delivery speed etc. No proof that it’s not at least as feasible that “the Covid + cheap money induced bubble” just started increasing in size faster… Because it’s just too painful of it wouldn’t.
> Startups are working faster than ever, and both businesses and consumers are demonstrating high willingness to pay for new products.
..I feel like the focus was more "offering (generative) AI features" than "built with AI", as in startups being AI forward for their ICPs are building businesses faster than the incumbents who are still trying to figure out how to wedge AI into their tech debt laden product landscape.
Even so, the increased rate of revenue generation cannot be explained by AI features since all the other (read: not skyrocketing) startups also offer AI features.
Did these portfolio companies just get the UX right? That would mean their entire portfolio, across industries, just happens to be filled with exceptional designers of products, who all simultaneosly cracked the UX for their respective industries.
A simpler inquiry:
What if the cause of this revenue growth is not on the startup/VC side but on the buyer side? Why are more companies rushing out there willing to spend more than before?
Here we have some candidates to consider: FOMO, labor efficiency, excitement.
Now, we have economics (behavioral and regular) to form a causal narrative; we have loss aversion and job cuts as incentives.
So, based on this post can we say this is a better time to start a startup than before? I don't think so.
Can we say that if you do start one with the same odds of success as before, grow it to a VC happy size, and have AI somewhere in the offering, you'll probably have more revenue than companies of a similar size from a few years ago? Yes, I think we can say that.
> Probably meant as a confidence boost to the sleep-deprived founders out there. But teaches nothing.
The post insist 2 to 4 million ARR in 1 year is the new norm. My guess its meant for their own investors and get founders to undervalue their achievements (Or learn to get creative with what ARR means).
I can’t speak to the world of startups or venture capital, I’m way too far from that ecosystem, but I’d like to add a perspective from the sidelines.
What stands out to me right now is just how loud the expectations around AI have become, especially among non-technical folks. It’s not just “Bitcoin hype” loud, it’s bordering on “AI will solve everything” levels of noise. For those of us who’ve been around a bit longer (sorry, younger HN crowd), the current buzz feels reminiscent of Y2K or the first dot-com wave.
Back then, I was early in my career, but I vividly remember the headlines, the overpromises, and the sheer volume of attention. The difference now is, there’s a lot more substance under the surface. The tools are genuinely useful, and the adoption curve feels more practical, even inevitable. That’s what makes me think AI might become to this era what the smartphone was to the last, not just a novelty, but an everyday dependency.
That said, I’ve also learned a lot from voices here on HN, especially when it comes to the financial realities behind the tech. If there’s one throughline in many of these discussions, it’s that financial viability, not just hype or innovation, is what ultimately determines whether this all collapses or truly transforms the world.
>financial viability, not just hype or innovation, is what ultimately determines whether this all collapses or truly transforms the world.
That is some of the best wisdom on HN. Beating your competition's AI model at whatever goalposts you think is important means nothing until you have positive cash flow. All hype must encounter reality and survive to not only make a sale, but then go and do it very consistently.
100% agree on this. It’s 1995 all over again. AI is as big (or bigger) as the Internet back then. Hype to totally insane levels but eventually all things that go up must come down.
In the meantime, the usual suspects are gonna make a whole lotta money.
The internet ended up just as big as predicted in 1995 (or bigger) - it just took a bit longer. What do we not have online today that was predicted in 1995?
I think discussions about AI hype miss a critical factor: there are two groups of people getting swept up in hype. One are the Investors[0]. The other are the Beneficiaries of the technology[1]. AI is over-hyped for the former, but not for the latter.
If AI hype is anything like dotcom boom - or like telecom, or building up railways in the US - well, it sucks for the Investors. For them, the hype is getting dangerous - if it's a bubble and it bursts, plenty of them will lose money, and many companies will fold.
But I'm not in that group, so I don't care.
For me, one of the Beneficiaries, the hype seems totally warranted. The capability is there, the possibilities are enormous, pace of advancement is staggering, and achieving them is realistic. If it takes a few years longer than the Investor group thinks - that's fine with us; it's only a problem for them.
--
[0] - In a broad sense, to include both people funding it and people making big investments around the expectations - whether regular investments, or company strategy, or career plans.
[1] - People using it for work and personally, researchers, etc.; also people with defined hopes for the technology; also ultimately everyone who benefits from it when it matures (and possibly builds on top of it).
> If it takes a few years longer than the Investor group thinks - that's fine with us; it's only a problem for them.
It is also for the beneficiaries because price comes into the equation and the longer it takes, the more expensive it will be.
We are currently paying the early-Uber prices at the moment but it's likely not sustainable (or not enough) and we'll see price hikes as soon as vendor lockin is sufficiently set in.
Assuming there remains more than one vendor, where the lock-in? I use Claude 4 via an IDE plug-in and both Claude and the plug-in (and the IDE!) are replaceable.
It is insane. You should see my inbox, daily links to AI articles, sales execs panicking about falling behind. Honestly, it’s understandable with all the noise, but it’s also hard to keep up.
I’ve lost count of how many times I’ve had to explain, again and again
“No, AI can’t do that… and no, it’s definitely not drawing up your architectural building plans.”
Well, not yet, anyway.
I lived through the dotcom boom too. It's a poor point of comparison, because as thrilling as the moment was, there weren't any techs then that could think or reason. And right now are we at the furthest point in its development? From the extreme pace of improvements it looks more like its infancy.
Yeah, I totally agree, we’re still at the beginning, and that’s what makes it a little scary
My dotcom comparison wasn’t really about the tech, more about the noise and hype. Feels like that same kind of frenzy, but now the tech’s actually capable of doing something big. The financial viability is still a big question though. Thanks for the comment.
Any S-curve looks like an exponential until it doesn’t. It’s impossible to make predictions like that. It’s in its infancy in terms of adoption all right.
I agree that financial viability is critical to the long-term prospects of a technology. It must deliver an ROI above other options. I'd recommend getting off the sidelines and jumping in to see what's happening. At the least, you'll have another perspective to inform your position. It's a pretty minimal investment to try it out.
You’re right to think that I probably do sound more like a cautious observer than I actually am. For what it’s worth, I’ve been experimenting with AI tools on the side (mostly in coding and writing workflows), and I’m planning to dive deeper soon, especially around integrating agents into my SaaS.
The post was more about the hype and attention surrounding AI, which can feel mentally exhausting at times, mostly because of how fast everything is moving. Not a complaint, really. If anything, that might be a good sign. I totally get why people are excited, it just takes effort to stay grounded in the middle of it all.
Appreciate the comment! Hopefully next time I’ll be jumping in with war stories instead of sideline takes.
The latest trend I've seen is blabber about AI super intelligence which will either kill us or lead to absolute utopia by 2030.
In the mean time, I try to enjoy the freely available LLMs for quick summaries on technical topics before the inevitable enshittification ruins them forever.
Eh... I honestly can't even pretend to see that far in the future.
What I will say is that AI will either fundamentally transform the economy for all of humanity, or it will fizzle out. I really don't see any in-between.
I think of it like Uber in the early days, when it was subsidizing rides to try to gain market share while ignoring taxi and labor laws until it could pay to change them. Uber's original plan was to bleed money until it could replace human drivers with robots, but that didn't work out.
The current AI companies are burning money with an exit strategy of replacing office workers with robots. If/when that doesn't happen, they'll have to jack up prices and figure out another business model. Uber had the two-sided market and network effects for a true enshittification play -- riders and drivers are both trapped -- but LLM companies haven't figured that part out yet. Do they go for ads once they have enough users and brand recognition? Hoard GPUs and training data (maybe through licensing deals) to create a moat?
TL;DR: there are two groups of people mixed up in the hype: the people investing in it, and people using it. AI may indeed be overhyped for the former. It's not overhyped for the latter.
Makes me think of how railways were built across the US. AFAIK, the first generation of investors generally lost big. They funded a huge, capital-expensive infrastructure project, and didn't get a return on it in time. But even as they lost, the work they funded remained - subsequent waves of businesses built on top of it and became profitable, the society benefited, and the country was transformed. The only losers to this "bubble" were the first-movers and their backers.
So when someone wonders if AI is overhyped, I'd ask them: what's your stake in this? Are you an investor hoping for quick returns, or are you someone who stands to benefit from the technology existing?
Totally agree that the tech can still be transformative even if investors lose money. The question for me is if it stops being financially viable, what keeps driving it forward?
> just how loud the expectations around AI have become, especially among non-technical folks.
This. It's bordering on mass madness. I am taking 2-4 calls a week from "two guys from ..." with mad ideas and unrealistic expectations of what it takes to build and maintain an AI product. I've seen it with early internet rush, Web 2.0, and crypto before.
The article lists the many ways in which the bar for success - the minimum "table stakes" that you have to achieve in order to be considered success - have drastically risen, and then concludes with:
> we believe there’s never been a better time to build an application-layer software company.
Nothing could be a clearer indication that the primary desirable quality in a founder is the conviction that, against all odds, you are better than everyone else.
> The article lists the many ways in which the bar for success - the minimum "table stakes" that you have to achieve in order to be considered success - have drastically risen
Have seen how application processes for technical roles went, in less than a year, from considering AI cheating; to now requiring AI to do the take home or finish a live coding task
Your loss, then. There is a reason that "shoot the messenger" attacks are referred to as ad hominem fallacies.
There may be loads I don't like about a16z, but this article contained lots of interesting insights and actual data. You don't have to agree with their conclusions to get tons of value out of the information they present.
Maybe it's the fact that we've seen VCs and a16 specifically push for highly overhyped things that didn't really come true like WeWork, Crypto?
If we go beyond a16, the last decade has shown us VCs pushing for insane valuations for companies that went under after hiring sprees. We've seen Mark Zuckerberg sing similar odes to Metaverse.
At some point you stop believing the next big thing from a certain crowd.
This doesn't mean I am not exploring how to use AI and even have something I am working on in terms of an idea, it's just that caution is advised.
In general, it’s unwise to take corporate comms at face value. Think about what Andreessen Horowitz would like investors and founders to believe and how it would like them to behave. Then consider the post as if it were an attempt instil those beliefs and elicit those behaviours.
This hides one major caveat in AI which is, contrasting to "old" tech, your OpEx on compute doesn't scale the same way your engineered app traditionally did.
In other terms, as your revenue scales, your OpEx scales too. This breaks the idea that you need to grow revenue to "break off" as your margin as set due to compute.
The other issue is, I've been burning compute between Perplexity, Grok, Gemini, Claude and Deepseek. I pay nothing for these and they are good enough. It is easy to grow revenue to $1m when you are burning $2m of compute.
Anyone else a bit confused by the use of the word "working" used, given the content of the post? I thought this was going to be about how white collar work is changing, not about fundraising and growth strategies.
This seems like a case of selection bias, where they are looking at all the Gen AI startups and seeing that they are making revenue faster than previous startups. But Gen AI startups have mostly only started very recently, so it's obvious that all the successes must have grown fast, as they haven't been around long enough to grow slowly. Maybe in 5 years, we'll see a lot of cases of successful startups that took a slower growth trajectory instead.
But whether it's short-sighted for the investors or not, I think the takeaway for founders is "investors now expect you to make more revenue faster, and B2C applications are more interesting than before".
I think this has an interesting consequence for founders and employees at startups in this day and age, and I'm not quite sure how I feel about it.
On one hand, it means you can "fail faster". That is, if you're a startup employee, and you don't see "hockey stick" growth that is looking crazy impressive at the end of year 1, you should know that the chances of your equity being worth more than a token are basically zero. Starting around the dot com boom, I worked in numerous startups, and for some of them we were still chugging along in years 3-4 with the hopes that our "semi-OK, decent growth" would turn vertical any day now. I've seen numerous startups that started in the 2015-2020 timeframe (so existed for 5-10 years) where they didn't outright fail but common got wiped in an acquisition. That's more a consequence of the rise in interest rates and difficult fundraising environment, but it's really rough to plug along at a company for 5-10 years, think you're doing OK, and then your stock is worth nothing. So from a startup founder/employee perspective, you get signal faster and don't have to waste time.
Simultaneously, though, it seems like any idea that would take a decent amount of upfront investment and time would be hella difficult to get funded, and I think that's unfortunate.
I read TFA. Quickly but I read it. It's short but I have no idea if they were saying "wow AI products are popular" or "AI helps startups reach higher levels of profitability faster" or simply "A company that says they are making an AI driven product receives more initial users and funding".
Each of those are so wildly different conclusions and require such wildly different data to support it.
The metric the article
focuses on is “revenue,” but it seems like the foundation many of these startups build on (other people’s LLM APIs) are much more expensive than the last generation of startups.
Given the cost of training a SOTA model, it’s not clear these companies have sustainable businesses. If your primary expense is AWS you can always shift to your own hardware once you hit sufficient scale. If you’re Cursor, how big do you need to get to eliminate your 3rd party API dependency?
The cost of training a good model can be as low as $5.5m (DeepSeek v3) or "a few tens of millions of dollars" (Claude 3.7 Sonnet https://simonwillison.net/2025/Mar/2/ethan-mollick/) so once you have proven revenue against other models it might actually become a reasonable thing to do.
This article is about AI applications, not core foundation models. None of the initial 3 companies mentioned (Cursor, Lovable, Gamma) train there own models from scratch, nor do they need to. E.g. I get tons of value from Cursor, but I also still pay for ChatGPT Plus.
There is also enough competition in the core model space that these apps don't need to have an Achilles heel by being reliant on a single vendor. E.g. I think Cursor was smart to let you "bring your own API key".
I get that the article is about applications, but the applications have a dependency on 3rd party models, and that dependency is costing them most (all? More than all?) of their revenue.
Put another way, if I make $100M in annual revenue, but am paying out $110M to the API I wrap, it’s not nearly as compelling a business as that top-line $100M number makes it out to be.
In the previous generation of startups, expenses were mostly dominated by headcount, and the cost of actually delivering the service tended to be small. The story was “keep growing revenue, and if you need to show a profit, stop hiring.”
An AI startup built on other people’s models has to hope that the foundational models end up being fungible commodities, otherwise any margins you might gain will get squeezed out by your LLM provider. Alternatively, you can train your own model.
I don’t know what Cursor’s userbase looks like. If everyone is paying for Pro but using their own API key, that’s obviously a high margin business.
Revenue only covers money flowing in. Virtually all companies in an extreme hype race like this one is deliberately losing money total (ie expenses are higher than revenue). Even during shit-hype times like food delivery apps, the companies were massively unprofitable.
I tried using Claude Code, was utterly disappointed with using it and it cost me a lot of money very quickly.
Just goes to show these tools can augment and improve your workflow and knowledge, but in their current state they cannot replace you - any CEO who says otherwise has no clue and is riding the hype train
Nowadays, I don't think work should be defined just by clocking in and clocking out. It’s about the ability to complete tasks and hit goals efficiently. This is how AI will redefine work.
If we want a feature we can write a two sentence prompt and get that feature. But the technical debt is going to grow exponentially, and I haven't seen a shred of focus on preventing that inevitable outcome.
The main claim in the post: Their portfolio companies have shown an improved rate of accumulating revenue ever since LLMs took off.
Weakest part of the post: No attempt at explaining how or why a LLM affects these numbers. They allude to 'shipping speed' and 'product iteration', but how an LLM helps these functions is left unexplored.
There's an implied deductive argument that a LLM can write some code, so obviously shipping speed is faster, so obviously revenue is faster. But the argument is never explored for magnitude of effect or defended against examples where shipping faster or using LLMs doesn't equal faster revenue.
Also, nothing about sampling bias, size or spread.
Overall: Probably meant as a confidence boost to the sleep-deprived founders out there. But teaches nothing.
> Startups are working faster than ever, and both businesses and consumers are demonstrating high willingness to pay for new products.
..I feel like the focus was more "offering (generative) AI features" than "built with AI", as in startups being AI forward for their ICPs are building businesses faster than the incumbents who are still trying to figure out how to wedge AI into their tech debt laden product landscape.
Did these portfolio companies just get the UX right? That would mean their entire portfolio, across industries, just happens to be filled with exceptional designers of products, who all simultaneosly cracked the UX for their respective industries.
A simpler inquiry:
What if the cause of this revenue growth is not on the startup/VC side but on the buyer side? Why are more companies rushing out there willing to spend more than before?
Here we have some candidates to consider: FOMO, labor efficiency, excitement.
Now, we have economics (behavioral and regular) to form a causal narrative; we have loss aversion and job cuts as incentives.
So, based on this post can we say this is a better time to start a startup than before? I don't think so.
Can we say that if you do start one with the same odds of success as before, grow it to a VC happy size, and have AI somewhere in the offering, you'll probably have more revenue than companies of a similar size from a few years ago? Yes, I think we can say that.
The post insist 2 to 4 million ARR in 1 year is the new norm. My guess its meant for their own investors and get founders to undervalue their achievements (Or learn to get creative with what ARR means).
What stands out to me right now is just how loud the expectations around AI have become, especially among non-technical folks. It’s not just “Bitcoin hype” loud, it’s bordering on “AI will solve everything” levels of noise. For those of us who’ve been around a bit longer (sorry, younger HN crowd), the current buzz feels reminiscent of Y2K or the first dot-com wave.
Back then, I was early in my career, but I vividly remember the headlines, the overpromises, and the sheer volume of attention. The difference now is, there’s a lot more substance under the surface. The tools are genuinely useful, and the adoption curve feels more practical, even inevitable. That’s what makes me think AI might become to this era what the smartphone was to the last, not just a novelty, but an everyday dependency.
That said, I’ve also learned a lot from voices here on HN, especially when it comes to the financial realities behind the tech. If there’s one throughline in many of these discussions, it’s that financial viability, not just hype or innovation, is what ultimately determines whether this all collapses or truly transforms the world.
Just my 2 cents.
That is some of the best wisdom on HN. Beating your competition's AI model at whatever goalposts you think is important means nothing until you have positive cash flow. All hype must encounter reality and survive to not only make a sale, but then go and do it very consistently.
In the meantime, the usual suspects are gonna make a whole lotta money.
I think discussions about AI hype miss a critical factor: there are two groups of people getting swept up in hype. One are the Investors[0]. The other are the Beneficiaries of the technology[1]. AI is over-hyped for the former, but not for the latter.
If AI hype is anything like dotcom boom - or like telecom, or building up railways in the US - well, it sucks for the Investors. For them, the hype is getting dangerous - if it's a bubble and it bursts, plenty of them will lose money, and many companies will fold.
But I'm not in that group, so I don't care.
For me, one of the Beneficiaries, the hype seems totally warranted. The capability is there, the possibilities are enormous, pace of advancement is staggering, and achieving them is realistic. If it takes a few years longer than the Investor group thinks - that's fine with us; it's only a problem for them.
--
[0] - In a broad sense, to include both people funding it and people making big investments around the expectations - whether regular investments, or company strategy, or career plans.
[1] - People using it for work and personally, researchers, etc.; also people with defined hopes for the technology; also ultimately everyone who benefits from it when it matures (and possibly builds on top of it).
It is also for the beneficiaries because price comes into the equation and the longer it takes, the more expensive it will be.
We are currently paying the early-Uber prices at the moment but it's likely not sustainable (or not enough) and we'll see price hikes as soon as vendor lockin is sufficiently set in.
I’ve lost count of how many times I’ve had to explain, again and again “No, AI can’t do that… and no, it’s definitely not drawing up your architectural building plans.” Well, not yet, anyway.
My dotcom comparison wasn’t really about the tech, more about the noise and hype. Feels like that same kind of frenzy, but now the tech’s actually capable of doing something big. The financial viability is still a big question though. Thanks for the comment.
The post was more about the hype and attention surrounding AI, which can feel mentally exhausting at times, mostly because of how fast everything is moving. Not a complaint, really. If anything, that might be a good sign. I totally get why people are excited, it just takes effort to stay grounded in the middle of it all.
Appreciate the comment! Hopefully next time I’ll be jumping in with war stories instead of sideline takes.
In the mean time, I try to enjoy the freely available LLMs for quick summaries on technical topics before the inevitable enshittification ruins them forever.
What I will say is that AI will either fundamentally transform the economy for all of humanity, or it will fizzle out. I really don't see any in-between.
The current AI companies are burning money with an exit strategy of replacing office workers with robots. If/when that doesn't happen, they'll have to jack up prices and figure out another business model. Uber had the two-sided market and network effects for a true enshittification play -- riders and drivers are both trapped -- but LLM companies haven't figured that part out yet. Do they go for ads once they have enough users and brand recognition? Hoard GPUs and training data (maybe through licensing deals) to create a moat?
Anyways, it's fun while it lasts.
https://news.ycombinator.com/item?id=44208831
TL;DR: there are two groups of people mixed up in the hype: the people investing in it, and people using it. AI may indeed be overhyped for the former. It's not overhyped for the latter.
Makes me think of how railways were built across the US. AFAIK, the first generation of investors generally lost big. They funded a huge, capital-expensive infrastructure project, and didn't get a return on it in time. But even as they lost, the work they funded remained - subsequent waves of businesses built on top of it and became profitable, the society benefited, and the country was transformed. The only losers to this "bubble" were the first-movers and their backers.
So when someone wonders if AI is overhyped, I'd ask them: what's your stake in this? Are you an investor hoping for quick returns, or are you someone who stands to benefit from the technology existing?
This. It's bordering on mass madness. I am taking 2-4 calls a week from "two guys from ..." with mad ideas and unrealistic expectations of what it takes to build and maintain an AI product. I've seen it with early internet rush, Web 2.0, and crypto before.
> we believe there’s never been a better time to build an application-layer software company.
Nothing could be a clearer indication that the primary desirable quality in a founder is the conviction that, against all odds, you are better than everyone else.
Have seen how application processes for technical roles went, in less than a year, from considering AI cheating; to now requiring AI to do the take home or finish a live coding task
There may be loads I don't like about a16z, but this article contained lots of interesting insights and actual data. You don't have to agree with their conclusions to get tons of value out of the information they present.
If we go beyond a16, the last decade has shown us VCs pushing for insane valuations for companies that went under after hiring sprees. We've seen Mark Zuckerberg sing similar odes to Metaverse.
At some point you stop believing the next big thing from a certain crowd.
This doesn't mean I am not exploring how to use AI and even have something I am working on in terms of an idea, it's just that caution is advised.
Reading your enemy’s propaganda may be useful sometimes, but taking it seriously is foolish.
In other terms, as your revenue scales, your OpEx scales too. This breaks the idea that you need to grow revenue to "break off" as your margin as set due to compute.
The other issue is, I've been burning compute between Perplexity, Grok, Gemini, Claude and Deepseek. I pay nothing for these and they are good enough. It is easy to grow revenue to $1m when you are burning $2m of compute.
But whether it's short-sighted for the investors or not, I think the takeaway for founders is "investors now expect you to make more revenue faster, and B2C applications are more interesting than before".
On one hand, it means you can "fail faster". That is, if you're a startup employee, and you don't see "hockey stick" growth that is looking crazy impressive at the end of year 1, you should know that the chances of your equity being worth more than a token are basically zero. Starting around the dot com boom, I worked in numerous startups, and for some of them we were still chugging along in years 3-4 with the hopes that our "semi-OK, decent growth" would turn vertical any day now. I've seen numerous startups that started in the 2015-2020 timeframe (so existed for 5-10 years) where they didn't outright fail but common got wiped in an acquisition. That's more a consequence of the rise in interest rates and difficult fundraising environment, but it's really rough to plug along at a company for 5-10 years, think you're doing OK, and then your stock is worth nothing. So from a startup founder/employee perspective, you get signal faster and don't have to waste time.
Simultaneously, though, it seems like any idea that would take a decent amount of upfront investment and time would be hella difficult to get funded, and I think that's unfortunate.
This article even smells ... generative.
Each of those are so wildly different conclusions and require such wildly different data to support it.
Given the cost of training a SOTA model, it’s not clear these companies have sustainable businesses. If your primary expense is AWS you can always shift to your own hardware once you hit sufficient scale. If you’re Cursor, how big do you need to get to eliminate your 3rd party API dependency?
There is also enough competition in the core model space that these apps don't need to have an Achilles heel by being reliant on a single vendor. E.g. I think Cursor was smart to let you "bring your own API key".
Put another way, if I make $100M in annual revenue, but am paying out $110M to the API I wrap, it’s not nearly as compelling a business as that top-line $100M number makes it out to be.
In the previous generation of startups, expenses were mostly dominated by headcount, and the cost of actually delivering the service tended to be small. The story was “keep growing revenue, and if you need to show a profit, stop hiring.”
An AI startup built on other people’s models has to hope that the foundational models end up being fungible commodities, otherwise any margins you might gain will get squeezed out by your LLM provider. Alternatively, you can train your own model.
I don’t know what Cursor’s userbase looks like. If everyone is paying for Pro but using their own API key, that’s obviously a high margin business.
If we want a feature we can write a two sentence prompt and get that feature. But the technical debt is going to grow exponentially, and I haven't seen a shred of focus on preventing that inevitable outcome.
It really is about "How do you know your startup is 'working' (i.e. doing the right things to be successful) in this AI era".