Pinned Tweet
"Human plus AI is going to outperform pure AI, so the question is what part do you want to keep human. As a venture firm, what you decide not to automate sends a strong signal about what you believe your strengths are, and what you believe your job is" - @yoheinakajima
Yohei is General Partner and co-founder of @UntappedVC, a research-oriented VC firm, and one of the most brilliant individuals I've had the pleasure of speaking to for @JoinOdin's Going Solo series.
To give you an idea of his range, here are just three of the papers he has published in the last few years:
• Task-driven autonomous agent utilizing GPT-4 — also known as "Baby AGI", one of the first open source automonous agents he developed in 2023, followed by the "Pippin" digital being framework.
• The Shared Discovery Paradox: How a One-Answer Rule Turns Better Information into Worse Search — taking a game theory approach to understanding the dynamics of problems like early stage investing.
• Soft-to-Rigid: Shape Continuation as an Alternative Search for Packing Congruent Shapes — outlining his novel (and record-setting) "hardening" solution to mathematical shape packing problems.
He's also built an autonomous research lab (Distributed Discovery) and a wikipedia style platform for organising knowledge around source data (Epistemedia).
His intense curiousity, and the creativity with which he approaches trying to find solutions or paths to better understanding, are both laudable qualities for a venture capitalist.
Check out the episode here: blog.joinodin.com/p/unpolish…
For every $100 the seven megafunds invested between 2005 and 2023, only about $11–19 matched the market-wide distribution of capital.
So far in 2026, it's about $38.
This group has gone from a set of differentiated firms to something that looks more like a VC megacap index.
Data sourced from @harmonic_ai.
Fewer opportunities + more capital = enmeshment
Per round, and per dollar, the large venture firms are increasingly similar in their coverage.
They converge on the same companies as funding rounds progress, herding in more capital to drive prices up further while adding logos to their LP deck.
Unfortunately, the underlying businesses do not become significantly stronger and price quickly accelerates far beyond value.
Thanks to the good folks at @harmonic_ai for letting me analyse their data for this. More to come.
Fewer opportunities + more capital = enmeshment
Per round, and per dollar, the large venture firms are increasingly similar in their coverage.
They converge on the same companies as funding rounds progress, herding in more capital to drive prices up further while adding logos to their LP deck.
Unfortunately, the underlying businesses do not become significantly stronger and price quickly accelerates far beyond value.
Thanks to the good folks at @harmonic_ai for letting me analyse their data for this. More to come.
For a long time, concentration in VC remained fairly stable, as capital inflows were broadly distributed.
This changed in the last five years as exits dried up and the "illiquidity ratchet" kicked in — only the largest and least sensitive LPs have sustained allocation, to the largest firms.
(i.e. Smaller LPs need distributions to recycle into new comittments, and therefore rely on the market acting rationally to produce exits. On the other hand, large firms produce irrational behaviour.)
The HHI (Herfindahl-Hirschman Index) is a standard measure of market concentration. It is the sum of the squared market share of each participant.
A score of >1,800 designates a highly concentrated industry, trending toward oligopoly, which receives extra antitrust scrutiny on M&A.
This must also factor the "relationship-based" of venture capital, where firms are not independent actors operating in competition with each other.
So, it's possible to calculate a "similarity-adjusted" score which considers the level of enmeshment — how frequently the portfolio companies of a particular firm share the same set of investors.
Bu this measure, venture capital has clearly reached a point where concentration is problematic, surpassing the 2023 antitrust threshold and even the higher original threshold from 2010.
As the market becomes more concentrated, this manifests as conesensus investing and increasingly inflated prices for mediocre assets.
The eventual "buyer" is yet more downstream private capital, each getting marked up by the next. There's no bagholder if the bag keeps getting passed around.
It's unclear how this is corrected without some form of intervention.
Dan Gray retweeted
Which processes should you look to automate, as a VC?
Decision making, like scoring opportunities, should probably be the last place you look for efficiency.
Start with the functions where you may otherwise be tempted to bring in an intern. Treat AI as a "cybernetic teammate", allowing you to focus more on the fundamentally human elements of venture capital.
"What you decide not to automate sends a strong signal about what you believe your strengths are."
Wisdom from @yoheinakajima, on this episode of Going Solo.
Yohei Nakajima came to venture as a builder.
He has made tools as a hobby since high school, without ever having held an engineering job, and he began sharing them with the world during COVID.
Today, @yoheinakajima is General Partner and co-founder of @UntappedVC, a small firm on the frontier of applying AI to venture capital.
His route there began about ten years ago at TechStars, helping spin up the Disney Accelerator, and runs back to a college thesis on angel investment organisations.
In this episode of Going Solo, he joined @credistick to discuss:
0:53 - Intro to Yohei and Untapped
3:33 - Should venture capital be more innovative
7:20 - Being a builder/hacker and a VC
10:40 - Applying AI to venture capital
14:40 - Mapping market data
17:06 - "Shared discovery paradox"
22:10 - VC bifurcation and the need for clarity
26:05 - AI and the influence on creativity
30:14 - Automation creeping earlier
33:35 - AI generated pitch materials
38:54 - The importance of follow-through
40:10 - The world in 10 years
The OpenAI and Anthropic IPOs are critical to this era of venture capital; the market needs to understand whether or not SpaceX was an anomaly.
The two private labs have grown to obscene valuations for private companies, but always in the shadow of SpaceX. In some ways, that offered legitimacy.
Ultimately, if they demonstrate that multiple companies can list at $1T+ valuations, and stick it, then there's some hope for the megafund model of VC.
If not, LPs may be deep underwater already as the pace of fund expansion has accelerated far beyond the general rate of growth in VC exit value.
Dan Gray retweeted
We got people saying you need $1B+ to do Seed investing
Meanwhile @credistick explains how your risk can actually go up if you raise too much money
Half of venture says capital is a moat.
The other half says huge rounds make your startup more risky.
@credistick on what you really sign up for when a mega-fund writes a big check:
"Their value prop, part of it at least, is that they will be less price sensitive when they invest. So they will give you much more money on much more friendly terms potentially.
The downside of that is the growth that you kind of obligated to deliver as a part of getting that huge check is way higher. The stakes of failure are higher. The risk you're taking on as a company by accepting that much money is much greater.
What should happen through a startup's life is as they develop, the risk slowly goes down. You end up raising these huge rounds and your risk shoots back up to the top again, and then you need to work it down again.
For some companies in some categories, it's the right move.
But for many it's probably not."
“Raise what the business actually needs, at a valuation it can grow into, and you keep good [exit] options open.”
Good read from @MorgLiv.
Bottom line: beyond a certain point, there are diminishing returns to raising capital.
Large rounds are a burden on the business, and make landing an exit harder unless they fund important milestones.
The idea that you can secure a business filling a bank account with VC dollars is madness; It is literally risk-seeking capital.
I’m certain this has nothing to do with Fidelity being an investor in both OpenAI and Anthropic.
We’re expanding access to the IPO market, so that more of our customers can participate. Starting today, October 19, we’re reducing the eligibility minimum to participate in IPOs, Follow Ons and Secondary offerings to $25,000 in retail assets (e.g., brokerage accounts, etc.). Learn more or sign up for alerts: go.fidelity.com/0aLSU8
Dan Gray retweeted
as a big fan of @credistick, was a pleasure to join him to chat about my thoughts on AI in VC and startups!
0:53 - Intro to Yohei and Untapped
3:33 - Should venture capital be more innovative
7:20 - Being a builder/hacker and a VC
10:40 - Applying AI to venture capital
14:40 - Mapping market data
17:06 - "Shared discovery paradox"
22:10 - VC bifurcation and the need for clarity
26:05 - AI and the influence on creativity
30:14 - Automation creeping earlier
33:35 - AI generated pitch materials
38:54 - The importance of follow-through
40:10 - The world in 10 years
btw if you're in SF, i'm doing a deeper dive into my agentic set up this Wednesday over a Lunch & Learn: luma.com/aiwithyohei
In a world where every pitch is stronger and more polished, because of AI, how do you make sure you can still identify the best founders?
You have to look at the "unpolished signals" from the founder; the data that hasn't been prepared for an investor — like their activity online.
@yoheinakajima, of @UntappedVC, has thought more about discovery, signalling and AI than most GPs.
Our conversation covered how he thinks about introducing AI into VC processes, and how VC processes must change in a world with AI.
Dan Gray retweeted
Yohei Nakajima came to venture as a builder.
He has made tools as a hobby since high school, without ever having held an engineering job, and he began sharing them with the world during COVID.
Today, @yoheinakajima is General Partner and co-founder of @UntappedVC, a small firm on the frontier of applying AI to venture capital.
His route there began about ten years ago at TechStars, helping spin up the Disney Accelerator, and runs back to a college thesis on angel investment organisations.
In this episode of Going Solo, he joined @credistick to discuss:
0:53 - Intro to Yohei and Untapped
3:33 - Should venture capital be more innovative
7:20 - Being a builder/hacker and a VC
10:40 - Applying AI to venture capital
14:40 - Mapping market data
17:06 - "Shared discovery paradox"
22:10 - VC bifurcation and the need for clarity
26:05 - AI and the influence on creativity
30:14 - Automation creeping earlier
33:35 - AI generated pitch materials
38:54 - The importance of follow-through
40:10 - The world in 10 years
In a world where every pitch is stronger and more polished, because of AI, how do you make sure you can still identify the best founders?
You have to look at the "unpolished signals" from the founder; the data that hasn't been prepared for an investor — like their activity online.
@yoheinakajima, of @UntappedVC, has thought more about discovery, signalling and AI than most GPs.
Our conversation covered how he thinks about introducing AI into VC processes, and how VC processes must change in a world with AI.
Yohei Nakajima came to venture as a builder.
He has made tools as a hobby since high school, without ever having held an engineering job, and he began sharing them with the world during COVID.
Today, @yoheinakajima is General Partner and co-founder of @UntappedVC, a small firm on the frontier of applying AI to venture capital.
His route there began about ten years ago at TechStars, helping spin up the Disney Accelerator, and runs back to a college thesis on angel investment organisations.
In this episode of Going Solo, he joined @credistick to discuss:
0:53 - Intro to Yohei and Untapped
3:33 - Should venture capital be more innovative
7:20 - Being a builder/hacker and a VC
10:40 - Applying AI to venture capital
14:40 - Mapping market data
17:06 - "Shared discovery paradox"
22:10 - VC bifurcation and the need for clarity
26:05 - AI and the influence on creativity
30:14 - Automation creeping earlier
33:35 - AI generated pitch materials
38:54 - The importance of follow-through
40:10 - The world in 10 years
"In hindsight, investing in Valar — pretty good move. At the time though, a lot of people looked at Isaiah and they thought he was delusional."
Many such cases.
Valar Atomics
30 days of Hard Tech: Day 8 ☢️
Abe Murray (GP at AlleyCorp):
“Isaiah is so phenomenal at telling a story, pulling a cult of belief behind him around what he's trying to do with nuclear power”
“Isaiah said from day one, ‘You have to touch metal. You have to build. You can't theorize your way to a new nuclear reactor. You have to actually do it.’”
“He understood what the world needs is cheap energy. It doesn't need a theoretically optimal reactor. It needs a buildable, scalable reactor, and that's what he's off doing.”
“And at the end of the day, Kevin Ryan (founder of AlleyCorp) asked me, ‘Is this one of the most incredible humans you've met?’ And the answer is yes, of course, and we backed him”
For a long time, concentration in VC remained fairly stable, as capital inflows were broadly distributed.
This changed in the last five years as exits dried up and the "illiquidity ratchet" kicked in — only the largest and least sensitive LPs have sustained allocation, to the largest firms.
(i.e. Smaller LPs need distributions to recycle into new comittments, and therefore rely on the market acting rationally to produce exits. On the other hand, large firms produce irrational behaviour.)
The HHI (Herfindahl-Hirschman Index) is a standard measure of market concentration. It is the sum of the squared market share of each participant.
A score of >1,800 designates a highly concentrated industry, trending toward oligopoly, which receives extra antitrust scrutiny on M&A.
This must also factor the "relationship-based" of venture capital, where firms are not independent actors operating in competition with each other.
So, it's possible to calculate a "similarity-adjusted" score which considers the level of enmeshment — how frequently the portfolio companies of a particular firm share the same set of investors.
Bu this measure, venture capital has clearly reached a point where concentration is problematic, surpassing the 2023 antitrust threshold and even the higher original threshold from 2010.
As the market becomes more concentrated, this manifests as conesensus investing and increasingly inflated prices for mediocre assets.
The eventual "buyer" is yet more downstream private capital, each getting marked up by the next. There's no bagholder if the bag keeps getting passed around.
It's unclear how this is corrected without some form of intervention.
There are five ways that the large venture platforms are trampling small firms:
1) Pumping price-insensitive capital into Seed, indiscriminately inflating prices / diluting outcomes.
2) Delaying liquidity, making it harder for smaller LPs to recycle capital into new commitments.
3) Changing deal terms to eliminate pro rata, take senior preferred or veto exits that are too small.
4) Obliging companies stay private for longer, eroding cash returns via a higher cost of capital.
5) Backing emerging managers for dealflow, cutting them off if they seem competitive.
This is a procyclical spiral, which results in small firms having a harder time raising capital which narrows the market and cedes more ground to the platforms.
This is problematic for the health of the venture market, as it ensures a continued march towards oligopoly. Competition is a force for good.
The deeper problem is that large venture platforms are structurally consensus oriented. They cannot do the job that venture capital was created for.
This is an empirical truth, evident in their portfolios. They are incentivised to dominate "legible" categories, not to look for new ideas on the frontier, and that is exactly how these firms are built.
Thus, the small firms which make up venture capital's discovery layer are gradually being pushed out of the market. The consequences of this, financially and economically, will play out over the next decade.
Guess what happens when the funnel of opportunities narrows? Prices rise further, capital concentrates, and advantages accrue to the larger firms.
Progress will stagnate, returns will fall, and the big firms will keep getting bigger. A perfect re-run of the 2010-2022 era, but perhaps much worse.
This cycle explains why firms like @usv and @benchmark have raised such large recent funds.
It's not that they are FOMOing in at the top of the market; it's more likely that they recognise the potential to be consumed by this spiral if they don't build a warchest to survive it.
Dan Gray retweeted
I've never bought art over €1,000 (not enough cash). But since I was a kid, my dad - a mathematician - has copied the painters he loves, just to figure out their techniques. He never sells anything (that's his behind me in the video). Because of him, I make art myself, and I co-initiated an art gallery/community, super-super-markt.com, I go and look at what's going on with curiosity.
I am not an expert, and I've never had money to collect. So what. You get into art by being in it - making, showing, arguing about it. Not by buying it.
Angel investing is the same. It looks like a rich person's sport from the outside. It isn't imo. Curiosity --> Obsession --> Suddenly, you're in.
thx @credistick + @JoinOdin !
When it comes to understanding how managers build conviction in early-stage companies, there are striking parallels between venture capital and art collection.
@enricomellis shared how angels and VCs can learn from the wisdom of art collector* Erling Kagge, in our recent episode of Going Solo.
(*also: explorer, philosopher, entrepreneur, and lawyer.)
Dan Gray retweeted
More than any time in my career, VCs seem to be underwriting other investors’ interest rather than the company itself.
Dan Gray retweeted
Here an allocator talks about "managers with good returns but without capital" why would a VC not have money if they are good at investing? As explained earlier in the thread, because allocators prioritize personal relationships and "trust" over returns xitter.cf/tseides/status/1608533…
Dan Gray retweeted
The large capital players in the private markets functionally have no opportunity cost on any given dollar, LPs reload them nearly automatically. That explains a lot.
“We found that growth mindset players performed better than fixed mindset players, their mistakes did not affect their attention to the game, and they read more learning feedback than fixed mindset players. In addition, growth mindset players were more likely to actively seek difficult challenges, which is often essential to self-directed learning.”
Gaming Mindsets: Implicit Theories in Serious Game Learning (2012)
In the translation of competitive gaming studies over to entrepreneurship, a clear attribute of interest is the growth mindset.
This refers to founders with a clear long-term objective, who recognise the journey there will involve constant learning and incremental achievments.
These aren't "strivers" chasing a finite victory. They are mission driven, open-minded, and positive-sum in their ambition.
