> Fortunes were made by people who didn't even understand their own tech
To be fair, do Larry and Sergey "understand" all of the tech that power myriad of Google's products & infrastructure? Often, folks in charge of rapidly growing businesses (innovations with disruptive potential / products with deep moats / upstarts riding hype cycles / moonshots on verge of breakthroughs) need to be also good at allocating the capital made available to them, including hiring prolific talent required by their respective ventures.
Worth looking at Jeff Dean's "Software Systems at Google" talk (2010) to understand what the tech at Google looked like before/after getting serious about scaling and maintainability.
There are so many funding options and average is hard to quantify here. But, if you really want the Unicorn path YC is one of the key Brands. If you want to build to something moderate (<100M ARR) then I think try one of the dozens of others. And even smaller plays can look at Angels, who've been writing more (smaller) checks.
The way I see it, moats are disappearing, and so pumping in more money to juice growth is not what it used to be - it’s not nearly as durable when nowadays people can switch on a dime, and the moment borrowing became a bit expensive, zillion dollar funding rounds started going away. So why give a chunk of a company to someone else?
I wonder how much the Section 174 changes to the tax laws that made software development less tax-advantaged contributed to this slow-down?
The 1st Trump Administration's big tax cut bill changed the tax laws for 2021 and on on how software development could be deducted, and in 2023 the IRS issued guidance for how to do that that made software engineering about 80% worse in terms of tax rates. The 2nd Trump Administration's big tax cut bill returned it to the old methods of deduction in 2025. So this bad tax situation for software development was in effect from 2022-2023 to 2025, which also lines up with the slow-down in funding.
> At Rebel, we score every YC startup and founder across hundreds of dimensions using our proprietary Rebel Theorem 5.0 ML/AI model, looking at everything from founder education and work backgrounds, personality traits, archetypes, entrepreneurial track record, founder-product fit, and more.
How would you prefer investment decisions to be made? Looks like there's a market for quant analysis where LPs respect what they're doing and companies are good to have them as investors.
We do stuff differently at Redbud Trail Group in alignment with what our LPs are looking for. VCs are just like any other two sided marketplace business trying to build a differentiated offering to both founders and LPs.
>How would you prefer investment decisions to be made?
Well, there's this thing in between everyone's ears that can make decisions based on what you might call intuition or the human factor.
Of course, to each their own. I personally wouldn't take money from someone who's only funding me because their algorithm says I passed its quantitative analysis.
My go to model would be for founders going to their local credit union and asking for a loan. The loan would be granted if the credit union members see that this thing brings any value for their local community.
VC funding has done too much damage, and it will continue to do damage no matter the method investors use to evaluate potential investments. I propose we give VC investments a break, say for one or two hundred years, at least.
How would you prefer investment decisions to be made?
Genuinely. Face to face. Without prejudice from a process that industrializes risk assessment and skips over nuance of the individual situation.
I guess that's the tradeoff if you're trying to scale this field, and why I'd prefer dealing with my investor or investee one on one. Even if it means that investment is delivered at a more incremental pace. Trust is a personal connection, forged over time.
They say they score the startups and founders but the startup or the product itself maybe fit under “and more”. They rate how determined the founders are to make money at any cost, not the startup, the idea, or the product.
I don’t understand pre series a investors that do this. If you need an algorithm to tell you a founder is good or not you are in precisely the wrong job.
I mean that sounds like the moneyball argument, saying an algorithm can't know more than the years of experience a scout carries with them. And then, surprise, the algorithm knew way more.
I would think some kind of unbiased AI would be better at picking out future successful founders than a VC, who could be much more easily mislead
And therein lies the root of the problem: it’s not unbiased.
AI systems have been shown over and over again to contain bias that reflects the biases in their training data.
And I’d bet it will favour some predictable combination of privileged background, Ivy League, yadda yadda yadda because those are the people life presents with the right opportunities at the right time.
But there will be huge numbers of other people that won’t be given the opportunity to succeed because the model will be blind to them, and that’s because the current system, from which the training data would be derived, is blind to them.
There's no ML model capable of reliably producing usable predictions from such high cardinality qualitative data so this is almost certainly just an LLM prompt + data scraped from LinkedIn or Pitchbook.
Even if we pretend it isn't just copywriting to scam LPs, the entire idea is that smart VC bets on founders should be made on the most public, surface level information rather than the subtle tells and cues that come up from diligence and interactions with founders.
In other words, the thesis is to just make decisions like a first year sourcing analyst rather than an experienced, talented Partner.
Isnt the point of entrepreneurship to blaze your own trail? Seems like machine learning algorithm trained on past entrepreneurs would inherently fail at identifying new ones.
The very nature of novelty is that it can't be computed ex ante. That goes for the entire causal chain leading up to the novel invention.
Any group doing this is forcing local optimization that eventually runs out, this is part of the mix that is killing innovation funding in the valley during the current cycle.
Investment is more lock step, less weird, than it has been in a long time, less diversity of diversity and more "capital always wins".
As the traunch of investors/founders becomes myopic and failure averse you'll no longer see companies do frantic life saving pivots either, the age of messy evolved startups is not likely this age.
These attempts are so deeply stupid. Data driven is so stupid, data informed is way more mathmatically sound.
In tech twitter there's a long running joke to explain frivolous things that occurred during the 2010s: "<insert subject here> was a low interest rate phenomenon."
The bar for series A was frequently just, do they have a product, is it growing, did they go to Harvard or Stanford.
People have no clue what a free for all the funding markets were. Fortunes were made by people who didn’t even understand their own tech
> Fortunes were made by people who didn't even understand their own tech
To be fair, do Larry and Sergey "understand" all of the tech that power myriad of Google's products & infrastructure? Often, folks in charge of rapidly growing businesses (innovations with disruptive potential / products with deep moats / upstarts riding hype cycles / moonshots on verge of breakthroughs) need to be also good at allocating the capital made available to them, including hiring prolific talent required by their respective ventures.
Worth looking at Jeff Dean's "Software Systems at Google" talk (2010) to understand what the tech at Google looked like before/after getting serious about scaling and maintainability.
This immediately got my interest. Sharing a link for anyone else who's interested [1].
[1] https://www.youtube.com/watch?v=modXC5IWTJI
No but they invented page rank and the early core systems
If you don't want to read the whole article here's the answer to the title:
"The sudden slowdown in YC startup Series A rates coincides with interest rates climbing through 2022 and 2023."
Thanks!
Will all of what's going on, is there much reason for the average startup to take VC funding anymore?
There are so many funding options and average is hard to quantify here. But, if you really want the Unicorn path YC is one of the key Brands. If you want to build to something moderate (<100M ARR) then I think try one of the dozens of others. And even smaller plays can look at Angels, who've been writing more (smaller) checks.
What would have made that change? AI maybe means hiring fewer developers but you’ve still got plenty of other costs. And AI costs.
More reason. With Dev skill becoming less important, the main advantage is the amount of capital you can deploy.
The way I see it, moats are disappearing, and so pumping in more money to juice growth is not what it used to be - it’s not nearly as durable when nowadays people can switch on a dime, and the moment borrowing became a bit expensive, zillion dollar funding rounds started going away. So why give a chunk of a company to someone else?
The moats allow you to get abnormal profit margin with lower effort.
If you don't have that, you just pump up a normal profit margin and try to take over the market place
Depends on what you're doing.
I wonder how much the Section 174 changes to the tax laws that made software development less tax-advantaged contributed to this slow-down?
The 1st Trump Administration's big tax cut bill changed the tax laws for 2021 and on on how software development could be deducted, and in 2023 the IRS issued guidance for how to do that that made software engineering about 80% worse in terms of tax rates. The 2nd Trump Administration's big tax cut bill returned it to the old methods of deduction in 2025. So this bad tax situation for software development was in effect from 2022-2023 to 2025, which also lines up with the slow-down in funding.
> At Rebel, we score every YC startup and founder across hundreds of dimensions using our proprietary Rebel Theorem 5.0 ML/AI model, looking at everything from founder education and work backgrounds, personality traits, archetypes, entrepreneurial track record, founder-product fit, and more.
Gross.
The whole idea behind the fund seems incredibly lazy.
But the name "Rebel Theorem 5.0 ML/AI model" makes me suspect they are full of shit.
Let's face it: It's Claude.
With a highly innovative proprietary prompt.
How would you prefer investment decisions to be made? Looks like there's a market for quant analysis where LPs respect what they're doing and companies are good to have them as investors.
We do stuff differently at Redbud Trail Group in alignment with what our LPs are looking for. VCs are just like any other two sided marketplace business trying to build a differentiated offering to both founders and LPs.
>How would you prefer investment decisions to be made?
Well, there's this thing in between everyone's ears that can make decisions based on what you might call intuition or the human factor.
Of course, to each their own. I personally wouldn't take money from someone who's only funding me because their algorithm says I passed its quantitative analysis.
That "only" is doing a lot of work... we don't know the extent of their full process.
My go to model would be for founders going to their local credit union and asking for a loan. The loan would be granted if the credit union members see that this thing brings any value for their local community.
VC funding has done too much damage, and it will continue to do damage no matter the method investors use to evaluate potential investments. I propose we give VC investments a break, say for one or two hundred years, at least.
It's hard for me to imagine decisions at that early a stage can be extensively data driven.
> How would you prefer investment decisions to be made?
The YC pitch and interviews seem like a pretty good system.
This rewards PT Barnum types who are better at hyping themselves.
How would you prefer investment decisions to be made?
Genuinely. Face to face. Without prejudice from a process that industrializes risk assessment and skips over nuance of the individual situation.
I guess that's the tradeoff if you're trying to scale this field, and why I'd prefer dealing with my investor or investee one on one. Even if it means that investment is delivered at a more incremental pace. Trust is a personal connection, forged over time.
Let's run it on the Rebel founder and team.
Yeah this feels like Halo Effect applied to a spec sheet based on common founder attributes
Goodhart's law needs to come for VC/startup funds
It's 100% lazy "Excel sheet driven" decision making, not a good measure for quality relative to ground truth
They say they score the startups and founders but the startup or the product itself maybe fit under “and more”. They rate how determined the founders are to make money at any cost, not the startup, the idea, or the product.
I don’t understand pre series a investors that do this. If you need an algorithm to tell you a founder is good or not you are in precisely the wrong job.
I mean that sounds like the moneyball argument, saying an algorithm can't know more than the years of experience a scout carries with them. And then, surprise, the algorithm knew way more.
I would think some kind of unbiased AI would be better at picking out future successful founders than a VC, who could be much more easily mislead
> unbiased AI
And therein lies the root of the problem: it’s not unbiased.
AI systems have been shown over and over again to contain bias that reflects the biases in their training data.
And I’d bet it will favour some predictable combination of privileged background, Ivy League, yadda yadda yadda because those are the people life presents with the right opportunities at the right time.
But there will be huge numbers of other people that won’t be given the opportunity to succeed because the model will be blind to them, and that’s because the current system, from which the training data would be derived, is blind to them.
So, yeah: gross.
There's no ML model capable of reliably producing usable predictions from such high cardinality qualitative data so this is almost certainly just an LLM prompt + data scraped from LinkedIn or Pitchbook.
Even if we pretend it isn't just copywriting to scam LPs, the entire idea is that smart VC bets on founders should be made on the most public, surface level information rather than the subtle tells and cues that come up from diligence and interactions with founders.
In other words, the thesis is to just make decisions like a first year sourcing analyst rather than an experienced, talented Partner.
Not brainpan diameter, birth sign, or favorite Power Ranger?
Weird that they'd be looking for a founder like Brian Armstrong or a team like Collision Brothers.
Actually, not weird.
Isnt the point of entrepreneurship to blaze your own trail? Seems like machine learning algorithm trained on past entrepreneurs would inherently fail at identifying new ones.
The very nature of novelty is that it can't be computed ex ante. That goes for the entire causal chain leading up to the novel invention.
Any group doing this is forcing local optimization that eventually runs out, this is part of the mix that is killing innovation funding in the valley during the current cycle.
Investment is more lock step, less weird, than it has been in a long time, less diversity of diversity and more "capital always wins".
As the traunch of investors/founders becomes myopic and failure averse you'll no longer see companies do frantic life saving pivots either, the age of messy evolved startups is not likely this age.
These attempts are so deeply stupid. Data driven is so stupid, data informed is way more mathmatically sound.
*all archaic/wrong spelling intentional.
Yucky. They took a good name like “Rebel” and made it lame! Shameeeeee
TLDR; interest rates
Innit. End of ZIRP. I'm sure the companies are just as good.
Are they? The batches are huge, trample all over each other, sell to each other, and honestly don’t seem as compelling to me personally.
ha, came her to post just that, down to every single character.
Even the semicolon instead of a colon?
yep. although i probably would have ended the sentence with a period.
That's a lot of words and a lot of flogging of proprietary scoring models to say "YC is a ZIRP"
"We often detect things that happen well in advance of other people."
"So here's this thing that happened 5 years ago."
In tech twitter there's a long running joke to explain frivolous things that occurred during the 2010s: "<insert subject here> was a low interest rate phenomenon."
> For this analysis, we used the same data to predict overall Series A rates by batch, and compared that to actual graduation rates.
Predicted series a rates by batch is a straight horizontal line. Truly groundbreaking.
Is Heyman's 10% better than others' 10%?