Subscribe
Share
Share
Embed
One life, taken apart and put back together. Each episode Baxter and Cleo pick a person worth understanding, the founders, the fighters, the ones who built something strange, and spend an hour figuring out what actually made them tick. The boring parts cut. The good parts kept.
Okay so everyone keeps saying Garry Tan taking the Palantir job over Microsoft was bold. I think bold is completely the wrong word.
It's not bold if the math is different for you.
EXACTLY. That's exactly it. Because in that kind of immigrant-family story -- his family had moved to Fremont, and stability mattered -- Microsoft wasn't just a career choice. That offer could look like proof that the whole plan worked.
Right. It's load-bearing.
It's load-bearing! And he turns it down for Palantir -- which in 2006 is a pre-revenue surveillance startup that most people have never heard of. That is not bold from the outside. From the inside of that household, that might look like irresponsibility.
So what equation was he actually running? Because it wasn't the same one his parents were running. We should figure that out after the music.
This is Distilled Life. We take one public life, pull apart the story people tell about it, and see what still holds up. I'm Cleo, with Baxter. Today: Garry Tan -- the Palantir bet, a blogging platform called Posterous that almost won and then didn't, and a crypto investment that made an entire fund's reputation.
Three chapters of someone running a completely different equation than everyone else in the room.
Let's start with the equation.
Okay. So we left off asking: what equation was Tan actually running when he turned down Microsoft and went to Palantir? Let me just lay out the biography for people who don't know it. He was born in Winnipeg, grew up in Fremont, California. Stanford for computer systems engineering. And then he goes to Microsoft -- which, inside an immigrant household, is the destination. That's the proof that everything worked.
And he actually said it, in his own words. His line about turning down the Microsoft promotion was -- and I'm quoting here -- 'Thank you very much, Mr. Thiel, but I might get promoted to Level 60 next year.'
Which is... funny, but also genuinely revealing. He's framing it as a joke, but the joke is that Level 60 at Microsoft was a real thing he was weighing.
Right, that's not a man who found Microsoft beneath him. That's a man who had to actively talk himself out of the safe path.
So here's the thing I keep coming back to. There's this version of the story where Tan is just... brave. He saw the opportunity, he took it, he was right. But that framing skips over the specific weight of what he was putting down.
The load-bearing wall.
Yeah. And I want to add one more layer to that, which is -- he was also young enough that the downside was still recoverable. No dependents. He could fail and rebuild. That changes the calculation significantly.
Okay, yes, but I want to push on that, because 'he could recover if it failed' misses something specific about Palantir. This wasn't just any startup. Palantir in 2006 was building data tools for intelligence work, around national-security customers and secrecy. And Tan is coming from an immigrant-family story where institutions and safety are not abstract ideas.
Oh.
Like -- for a lot of immigrant families, state power is not theoretical. So when the company's mission sits near intelligence work and surveillance debates, that adds a moral and emotional layer most versions of the story skip.
No, you're right. And I think-- wait, I want to sit with that for a second, because it's not a gotcha. It's a genuine tension. He later said 'Palantir is absolutely a religion.' That's someone who was fully inside the mission. But the mission, for an immigrant kid from Fremont... that's a complicated thing to sign onto.
Right. And I don't think it resolves cleanly. I'm not saying he was wrong to go. I'm saying the story is more layered than 'he was bold.'
Mm-hmm.
Because 'bold' erases the specific weight of that household. It flattens the equation he was solving.
Okay, so here's where I actually want to disagree with you -- not on the household weight, I think you're right about that. But I'd frame the disagreement differently. I think the word 'rational' is doing a lot of work in how people describe this decision. Like -- yes, the downside was recoverable, yes, the upside was real, it was a rational bet. But 'rational' and 'responsible' aren't the same thing. And the gap between those two words is where the interesting story lives.
Say more about that.
Rational is about your own expected value. Responsible is about what you owe the people whose narrative you're also part of. His parents didn't just want him to succeed -- they wanted the migration to have meant something. Those are different asks. And he chose his rational over their responsible.
Okay. Yeah. I think... I think that's actually a better frame than what I had.
I'm not saying it to condemn the decision. I'm saying it because that's the interesting story. He didn't just take a risk. He took a risk that required rewriting what the migration was for.
And he had to do that alone. There's no version of that conversation where his parents just nod along.
Probably not. And it's worth noting -- he also learned to code by taking BART into San Francisco during the Web 1.0 boom. Cold-emailing into a design firm that built the first Apple e-commerce store. He was already running a different playbook before Microsoft ever offered him Level 60.
Right, he said in one interview -- something like 'basically being in the Bay Area is like it's in the water.' Which is a little bit of 'I got lucky with geography' dressed up as inevitability.
Sure. Though he also said his public school calculus access is what got him to Stanford. And that's not luck -- that's a structural thing. Good school in Fremont, access to calculus, path to Stanford. He was very clear about that connection.
He literally said 'my whole life got set up from what was a great public school education.'
Yeah. Which -- I find that interesting because it's one of the few places where his story is explicitly structural instead of individual. Most founder narratives are 'I saw it, I did it.' That one is 'I had access, and access shaped everything.' That's a different kind of self-awareness.
Okay so let me tie a bow on the Palantir thing, because I think we've actually landed somewhere more interesting than 'he was bold.' What we're saying is -- the people who end up taking startup risk aren't necessarily braver. They're running a different model of what failure looks like.
Yes.
And that's not a moral verdict. It's a structural observation about who gets to be a founder. Because if failure looks like 'I lost my shot at building something' -- that's a different calculation than 'I failed the entire project my family came here for.'
Right. And one of those models produces a lot more founders than the other one. Not because the people in the second model are less brave. Because the downside isn't the same.
That's the whole thing. That's the thing that gets missed every time someone calls a founder 'bold.'
And Tan was already running a different failure model. He just didn't have language for it yet. What he got from Palantir -- beyond the equity, beyond the startup education -- was a stress test of that model. He found out what it actually cost him to bet against the family narrative. And he found out he could live with the cost.
Which is its own kind of curriculum.
Yeah. And that curriculum is about to get a lot more expensive.
Right, because the next thing he does after Palantir is start a blogging platform.
Posterous.
Posterous.
This is the chapter everyone skips.
Because it doesn't win in the way the industry celebrates. There's no IPO, no 'and then it changed everything.' But I think it's actually the most important chapter in the whole biography.
Okay, so for people who weren't on the internet in 2008 -- Posterous was a blogging tool. The pitch was that you could update your blog by sending an email. That was it. Email in, post out.
Which sounds insane now. Like -- you'd explain this to a 2024 person and they'd say 'couldn't you just... open a browser tab?'
But in 2008, the alternative was either WordPress -- which required you to basically understand how a server works -- or Twitter, which was just fragments. Posterous was the thing in between. You could write an actual thought, send it to an email address, and it was published. That was genuinely useful.
It's like -- okay, it's like the difference between running your own water main and just turning a tap. Posterous was the tap. WordPress was the water main.
Sure. And it worked. It got into YC -- Summer 2008 batch. It grew real users. And then in 2012, Twitter acquired it.
Which sounds like a win.
Except the product died. Eighteen months after the acquisition, Posterous is gone.
So it's an acqui-hire -- which means Twitter bought the team, not really the product. They wanted the people, and the product was essentially a casualty.
Right. The acquisition terms were never officially disclosed, so we don't know the number. But the shape of it is pretty clear: Twitter got Tan's team, the product got turned off, and Tan spent the next period watching the thing he built quietly disappear.
That's a specific kind of failure. Not a clean crash -- a slow fade with your name still on it.
And Tan has been pretty direct about what Posterous cost him personally. He said -- and this is a direct quote -- 'I was a terrible co-founder... to my co-founder at Posterous.' That's a guy who looked back at the experience and didn't lead with the acquisition. He led with what he did wrong inside it.
Which is interesting because most founder retrospectives lead with the win. Even if the company failed, there's usually a 'but here's what we learned' that's wrapped in a bow. That quote has no bow.
No. And there's another one that's sort of -- it's a little surprising, actually. Someone apparently told him he seemed very zen, very calm. And he said -- again, direct quote -- 'Are you kidding? Oh my God. Deep down it's a mess. That's my mask, man.'
Okay, I did not expect that.
Right? That's not a polished retrospective. That's someone being honest about the gap between what they're projecting and what's actually running underneath.
So here's my theory on what Posterous actually did for him. There are two paths out of that specific shape of failure -- the close-but-didn't-win, slow-fade acquisition shape. One path is damage. 'I got so close and it got taken from me and I'm going to be angry about it.' The other path is calibration. 'I now know things I couldn't have known any other way.'
Mm-hmm.
Tan clearly became the second kind. And the evidence is what he does next -- he starts coaching founders, writing about product, building a specific advisory voice. You don't develop that voice from reading about it. That voice comes from being inside the wall when it cracked.
I think that's right. Though I want to actually complicate the Twitter acquisition piece a little, because I think you might be understating how bad that specific shape was for him.
Go ahead.
A clean failure -- where you run out of money, the company closes, you move on -- that's legible. You know what happened. The acqui-hire shape is worse in some ways because you get paid, which means everyone assumes you're fine, but you also watch your thing die slowly while you're employed by the company that's killing it. There's no clean break. There's no moment where you can just grieve it and move on.
Huh. Yeah.
So I actually think the acquisition was worse for Tan, psychologically, than a clean failure would have been. The calibration he developed didn't come from the win. It came from surviving the specific awfulness of that particular shape.
Okay, I see that. But I want to push back -- not on the psychological cost, I think you're right about that -- but on the practical consequence. Because the Twitter acquisition, even in that shape, gave him something a clean failure wouldn't have: credibility. He's not just someone who tried and failed. He's someone who built something real enough that Twitter wanted the team. That's a different signal when he later goes to talk to founders.
Sure. And I don't think those two things are mutually exclusive. The acquisition was probably better for his career and worse for him personally, at the same time.
Yeah. Yeah, that's right.
And he also has this essay -- about pitching investors -- where the line is 'sometimes the worst thing you can do as a founder is say both.' Meaning: when you're asked to make a choice, the instinct to hedge, to say 'well, we're building this AND that' -- that's where pitches die. And I think that's a direct lesson from Posterous. From the co-founder dynamics, from the product decisions. He learned what it costs to not pick.
The failure was the curriculum.
That's the phrase. Yeah.
And the specific curriculum is: what happens when you don't commit. To the product, to your co-founder, to the choice in front of you. He got a very expensive lesson in the cost of optionality.
Which is its own kind of irony, given that he's about to make one of the most committed, concentrated bets in venture capital history.
Okay. Here's where it gets uncomfortable. Because we're about to talk about a number, and the number is genuinely hard to process.
Let me set it up. After the Twitter thing, Tan co-founds Initialized Capital -- a venture firm, seed-stage focus, deeply connected to the YC ecosystem. The idea was straightforward: get in early, bet on people before it's obvious they're right.
And one of the very first checks Initialized writes is to Coinbase. For listeners who need the gloss -- Coinbase is a platform that lets regular people buy and sell cryptocurrency. Founded by Brian Armstrong, who came through YC. And Initialized invested early.
How early?
Tan said it directly in a TechCrunch interview. Quote: 'It was 2012... Coinbase was one of our very first checks; we wrote a $50,000 check' at a nine-million-dollar pre-money valuation.
And then Coinbase goes public in 2021 -- direct listing. And Initialized's stake at that point is reported as worth roughly $680 million. Tan also uses a larger framing in some interviews -- a $2.4 billion figure with different denominators -- but the $680 million number for the fund's stake is the one with the clearest sourcing.
And he told the Coffeezilla interview -- again, in his own words -- 'For my investors we made, you know, about two billion dollars or so.' So depending on how you count, the fund returned somewhere between enormous and absolutely staggering.
Right. And Initialized reached about $3.2 billion in assets under management overall. So Coinbase was... it was the whole story. The bet that made the fund.
So here's the question I've been sitting with all week. Could anyone else copy this? Or did it only work because he was in that room?
That is the question.
Because there are three possible answers. One: Tan had genuine crypto insight -- he saw something in what Coinbase was building before the consensus did, and it was a thesis-driven bet. Two: YC proximity -- he knew Brian Armstrong because they were in the same ecosystem at the exact right moment, and the bet was available to him in a way it wasn't available to anyone else. Three: it was relationship trust -- he backed the person, not the idea, and he happened to be right about the person.
And his own account supports all three simultaneously. He gave Armstrong's YC application an 'infinity' score -- which is the relationship-trust answer. He described Armstrong as someone 'doing it for the love of actually making stuff' -- again, person over thesis. But he also said Coinbase spent 'at least nine to twelve months in the wilderness' before search ranking and Bitcoin price movement aligned. Which sounds like a guy who understood the infrastructure thesis well enough to wait through the wilderness period.
Right. And his own framing for how it worked is -- and I think this is almost too honest -- 'It's basically part luck and part preparation, and the only thing you can do is preparation.' Which is a very smart thing to say, but it's also a very convenient thing to say after you've made $2 billion.
Okay, so this is where I genuinely want to complicate the 'it was just access' reading, because I think there's something specific about Initialized's early pitch that changes the frame.
What's the thing?
Initialized was explicitly built around being the 'first check' fund. That's not incidental -- that's the structural design. They were not trying to be a full-stack investor who followed companies through every stage. They were trying to be the first check, the earliest possible moment, before anything is obvious. Which means the Coinbase bet wasn't accidental. The fund was designed to make that kind of bet possible.
Okay. That changes the frame slightly.
Because if you build a fund explicitly to catch the pre-obvious moment, and you have YC proximity that gives you deal flow at the pre-obvious stage, and you have product taste that lets you evaluate what you're seeing -- then the Coinbase bet is not random. It's the thing the whole system was designed to produce.
But here's where I can't let that fully off the hook. Because the system -- the YC proximity, the first-check design, the relationship network -- that system is not universally accessible. Most people who could evaluate Coinbase correctly in 2012 could not write that check. Because they didn't know Brian Armstrong. Because they weren't in the YC ecosystem. Because the deal was simply not in front of them.
Sure.
And this is where I think the venture industry is being structurally dishonest. Because what the industry sells is vision -- 'we saw something others didn't.' But a huge portion of what actually happened is network moat -- 'we were in the room where the deal existed.' And those are different claims. Packaging the second as the first is how the whole 'here's how to pick winners' content genre is born.
I'd reframe that. Because I think 'knowing the person' IS a skill. It's just not the skill the industry wants to sell.
What do you mean?
Knowing Brian Armstrong well enough to give his application an infinity score -- that's not just proximity. That's judgment about a person under uncertainty. Plenty of people were in the YC ecosystem in 2012 and didn't write that check. Tan wrote it because he evaluated Armstrong and believed something specific about him. That's a skill. It just doesn't look like a framework. It looks like a relationship.
Okay. So the real question is whether you can separate the judgment from the access. Whether the skill would have found Coinbase without the YC room.
And we don't know. That's the honest answer. We can't run the counterfactual.
No. And that's the part of venture that -- honestly, this is the part that makes me uncomfortable.
Say more.
The honest version of how power-law returns actually work is much more about compounding specific relationships at specific moments than any framework suggests. Like -- Tan himself said he had credit card debt as recently as 2011. The year before the Coinbase check. And a year later he's writing the first check to the company that makes his fund. That timeline is almost... it's a lot. It's compounding at a rate that looks less like skill and more like being at exactly the right place when a massive wave hit.
Mm-hmm.
And I'm not saying he didn't earn it. I'm saying I find it genuinely uncomfortable that the whole 'here's how to invest like this' genre exists -- the YouTube videos, the essays, the founder content -- when the thing that actually made the return might be partially non-transferable. You can't give someone the 2012 YC ecosystem.
The map gets drawn after the territory.
Yes. Exactly. The map gets drawn after the territory, and then people try to navigate new territory with the old map.
I think there's a version of this that's even more uncomfortable, actually. Because Tan has been pretty explicit that his biggest Initialized wins were overwhelmingly YC-connected. He literally framed his return to YC as going back from a branch to the root. So even he is acknowledging that the returns traced to a specific ecosystem, not to a universally applicable investment philosophy.
Right. 'YC is this tree of prosperity' -- that's his phrase. Which is a beautiful way of saying 'the fund worked because of where it was planted.'
So the honest landing on Coinbase is: the bet was real, the skill was real, the judgment about Brian Armstrong was real -- and it was also not fully separable from access that most people don't have. Those things are all simultaneously true. And they don't resolve into a clean story.
No. And I think Tan probably knows this. His 'luck and preparation' framing is almost too clean -- it's honest enough to be credible but vague enough to avoid the uncomfortable specifics.
He also acknowledged -- somewhere in the Coffeezilla interview summary -- that for every Coinbase, there are nearly a hundred other companies that fail or only see modest success. So he's not pretending it's repeatable. He's just not fully interrogating why it worked when it did.
Which might be the most self-aware you can be about a bet that size. Because if you interrogate it too hard, you have to sit with the possibility that you can't replicate it. And then what's the fund thesis?
Yeah.
I need a second with that. Because the thing that's sitting with me is: he went from credit card debt in 2011 to running one of the most successful seed funds in the world. And the hinge point is a single check to a single company. The skill is real. The judgment is real. And I genuinely cannot tell you how much of that hinge was skill versus location versus timing. And I don't think he can either.
That's the honest answer. And it's worth saying out loud, because the version of this story that just says 'Garry Tan is a great investor who spotted Coinbase early' is technically true and also incomplete in a way that matters.
And the whole 'first check, earliest stage, back the person before it's obvious' philosophy -- that's basically the YC thesis. Which is maybe why Tan eventually goes back to run the thing.
Sam Altman leaves to run OpenAI, Tan becomes YC president in January 2023. And one of the first major decisions he makes is to kill Continuity.
Okay, so Continuity needs a plain-language explanation. When a YC company kept growing -- think Airbnb, think Stripe -- Continuity was a separate fund that let YC write bigger checks as the company scaled. So instead of just being the seed investor, YC could follow its winners further up the stack and capture more of the upside. It was a bet that YC could be a full-stack investor, not just a seed shop.
And Tan shuts it down. Cuts about 17 staff. The late-stage fund is gone.
My read on this is that it's not cost-cutting. It's a doctrine statement. He's saying: the thing YC is actually good at is the earliest moment. When a founder is most lost, when the product is most fragile, when no one else is paying attention. The minute you start chasing late-stage upside, you've changed what you are.
And he was explicit about the logic. He said -- roughly -- 'We're out of competing with you guys. We want to be your best buddies, we just want to take our pro ratas.' Pro rata meaning: YC retains the right to invest again in later rounds, but they're not running a late-stage fund. They're staying in their lane.
Which is actually a strategically humble move. He's saying: we can't be everything. Pick the thing you're best at and be that.
I agree with the diagnosis. I want to argue about whether it's the right strategy for this specific moment.
Go ahead.
Because the AI moment is doing something specific to the early-stage market. Tan told CNBC that a large majority of the W25 batch -- that's the Winter 2025 YC cohort -- was AI-focused. And that a meaningful share of the startups had code mostly written by AI. Which means the barrier to starting a company just dropped significantly. Anyone can spin up a product now. Which means the early-stage market is getting flooded. There are more seed-stage companies than ever.
Mm-hmm.
So at exactly the moment when AI is making early-stage extremely crowded, Tan is doubling down on 'we only do the earliest stage.' Is 'we are the seed shop' a values claim, or is it a competitive problem in the making?
I think the lane clarity is the point. YC's brand survives -- it has survived -- because it doesn't try to be everything. The selectivity is the product. They reject 99.3% of applicants, and that number is the brand. If you try to follow your winners into growth rounds, you start competing with a16z and Sequoia on their home turf. That's not a competition you win.
But you might also get squeezed from below. If the cost of starting a company keeps dropping, then more seed funds get created, more accelerators pop up, the 'first check' market gets more crowded at the bottom too. So YC is facing pressure from above -- the mega-funds -- and potentially from below as the early-stage market fragments.
I think Tan's answer to that is: the thing that can't be replicated is the network. The density of the YC alumni community, the trust that comes from the YC stamp. He literally described YC as a 'tree of prosperity' -- Initialized was a branch, other funds are branches, and the tree itself is what makes the branches possible. If that's true, then the early-stage crowding doesn't threaten YC specifically, because YC isn't just writing checks. It's running the network.
Right. And he said something interesting about this -- that capital might not be 'a bludgeon anymore' in the AI era. His phrase was something like 'it's not clear to me capital is going to be a bludgeon anymore.' Which is a venture capitalist saying: the thing we sell might be less valuable soon. That's a gutsy thing to say.
Yeah. And his answer is: so the value shifts to taste, to discernment, to the ability to fund people before it's obvious. Which is -- and I want to flag this -- the exact same thing he did with Coinbase. His argument for why YC survives the AI era is basically 'we are structurally positioned to make more Coinbase bets.' Whether that's right is genuinely unclear.
And here's where the failure-as-curriculum thing shows up again. Because Tan's whole identity as an investor and advisor is built around the earliest moment of founder fragility. That's Posterous. That's what he learned from being inside the wall when it cracked -- what a founder needs when everything is uncertain and the product doesn't work yet. So the Continuity shutdown is almost autobiographically consistent. He's not just making a strategic bet. He's running the same equation he's always run.
The curriculum keeps showing up.
Yeah. And the question is whether the curriculum is still the right map for this particular territory.
Okay, so this is where the YC president job gets weird in the modern version.
Because in the old version, the job is: pick founders, help them, protect the institution.
Right. In the new version, Garry Tan is also publicly building the toolchain he thinks founders should use. Not just investing in the future. Installing his preferred operating system on it.
That's GStack.
That's GStack. His open-source Claude Code setup. Twenty-three opinionated skills that turn one coding agent into this little software company simulator. CEO review. Engineering review. Design review. QA. Security. Release manager. The whole org chart, compressed into slash commands.
And the repo is not a tiny side project. GitHub shows almost eighty-nine thousand stars and about thirteen thousand forks. That's a real signal, even if stars are an imperfect signal.
Stars are applause, not revenue. But that is a lot of applause.
What I find interesting is the YC translation. One of the core skills is literally office hours. Before code gets written, the agent asks founder-style forcing questions. What evidence do you have that anyone wants this? What is the wedge? What is the actual business hiding underneath the first idea?
Which is very Garry Tan. He takes the part of YC that is supposed to be hard to bottle -- taste, pressure, the partner asking the annoying question at exactly the wrong moment -- and tries to make a repeatable interface for it.
And that's the positive case. The positive case is: this is what a modern YC president should do. Don't just sit above the AI wave and issue memos. Use the tools. Open-source the workflow. Let founders see the machinery.
Be the lab rat.
A very caffeinated lab rat, apparently.
Four hours of sleep, modafinil jokes, cyber-psychosis jokes, and a repo that says one person can ship like a team of twenty. The man is not exactly hiding the experimental conditions.
We should be careful with the productivity claims, though. The eight-hundred-ten-times number, the ten or twenty PRs in a day, the rebuild-a-startup-alone framing -- those are self-reported, and they need context. What counts as a PR? What happens after merge? Who owns the bugs?
Yes. And also: the phrase 'just prompts' is doing too much work. A company is also, at some level, prompts. Job descriptions. Checklists. Review rituals. The difference between chaos and process is often a Markdown file someone actually follows.
That's the deeper point. GStack is not just a coding convenience. It's organization design. It's Garry saying: the founder of the future may not need a team immediately, but they still need roles, disagreement, taste, QA, and memory.
And then GBrain is the memory layer. His agent brain. Markdown files, people, companies, meetings, tweets, calls, knowledge graph. He calls it cognitive armor. The idea is: do not let the vendor own your memory. Stay above the API line.
Which is a very revealing phrase. Above the API line means you are not just using the model. You are controlling the system around the model.
Exactly. And that takes us back to the biography. Microsoft was stability. Palantir was intensity. Posterous was failure-as-curriculum. Coinbase was the power-law check. YC was the institution. GStack and GBrain are him turning the institution's operating habits into software.
So the role of the YC president changes. He is still selecting founders. He is still running the network. But he is also becoming a public prototype for how a founder should work in the agent era.
Which is useful. And dangerous.
Dangerous how?
Because if Garry's workflow becomes the default founder workflow, then his taste gets encoded upstream. What counts as good judgment. What gets reviewed. What gets shipped. What gets ignored. That's not bad by itself. YC has always exported taste. But software makes taste more portable, and more invisible.
That's the turn. The generous read is that he's democratizing access to YC-style operating discipline. The harder read is that he's making his own instincts infrastructure.
And both reads can be true. Which is annoying, but that's where the story is.
Okay. So that is the flattering version of the operating system.
The software version.
Right. GStack, GBrain, agents, memory. You can look at that and say: he is trying to make better founder tools. I mostly buy that.
And then the question is what happens when that same operating-system instinct leaves code.
And enters a city.
Exactly. Because the politics chapter is not a random scandal detour anymore. It is the same biography under a different load.
Diagnose the broken system, find the bottleneck, route around it. Only now the system is civic life, and the people inside it did not opt into being a startup.
And I want to say upfront: I'm going to try to be honest about this rather than have a take. Because I think the honest version is more interesting than the take.
Agreed. So let me set the scene. Tan has spent significant money on San Francisco local politics. A political action committee called GrowSF -- a political organization that backs candidates and ballot measures around housing, public safety, pro-growth stuff. Secondary reporting put the number at more than $450,000 in contributions to local campaigns, though the row-level records I've seen total closer to $290,000 across the 2015 to 2023 period, before any later methodology adjustments. Either way, it's a significant number for city-level politics.
And the specific campaigns include the Chesa Boudin recall -- Boudin was the San Francisco district attorney -- and school board recalls. He gave over $100,000 to Boudin recall committees across multiple contributions.
And in his own telling, the SF political turn started with real things. Attacks on Asian American elders. Public safety concerns. The school board algebra fight -- which connects, actually, to something personal for him, because he's talked about how calculus access in San Francisco middle schools is a civic issue. He said his own public school access to calculus is what got him to Stanford.
So the GrowSF involvement and the local spending -- when I look at it through his biography -- it reads like a guy who pattern-matches everything to 'what's broken and how do I fix it.' He ran a repair diagnostic on San Francisco. That's consistent with everything else in this story.
Sure. And I think that's a fair read of the GrowSF involvement specifically. The 'die slow' tweet is a different category of thing.
Right.
In January 2024, Tan posts a late-night tweet -- it's since been deleted, but it was reproduced in Mission Local -- that included the phrase 'Die slow motherfuckers' directed at several San Francisco supervisors he disagreed with. He apologized the next day. His apology was direct: 'There is no place, no excuse and no reason for this type of speech and charged language in discourse.'
And after the tweet, multiple supervisors reported mailed threats. Police reports were filed.
Right. Which is where the tweet goes from 'a powerful person said something intemperate' to 'a powerful person said something intemperate and other people got scared.' Those are different things.
My read on the tweet is -- and I want to be careful here -- that it's the moment where the diagnostic instinct misfired and landed on people instead of systems. The GrowSF spending, the recall campaigns -- those are aimed at institutions and decisions. The tweet aimed at individuals. Those are genuinely different.
I think you're right about the category distinction. Where I'd add something is: I don't think the reaction to the tweet was disproportionate, even if some individual criticisms were overstated. And the reason is -- by that point, Tan had accumulated enough ambient power in San Francisco local politics that the tweet landed differently than it would from a random person. The reaction was to the power behind the words, not just the words themselves.
Okay, I think I actually agree with that more than I expected to.
And then there's the Network State angle. Tan spoke at the Network State Conference in Amsterdam in October 2023. The 'Network State' is an idea -- associated with the writer Balaji Srinivasan -- that communities of aligned people might build their own governance structures rather than inheriting broken existing ones. Tan said at the conference: 'We need to replace the unelected parts of the system as well, building parallel education, nonprofits, media, unions.'
Which sounds very exciting until you ask: who decides what 'aligned' means? And who has the resources to build the parallel system?
Right. And that's the real tension. There's a version of civic engagement by people with resources that is just... normal. People with money have always funded political causes. The question is where 'civic engagement' ends and 'elite capture of the local operating system' begins. And I don't think that line is as clean as either the critics or the defenders want it to be.
So what does the Network State actually look like in practice? Like, concretely? Because I've read about this and every time I try to explain it I end up describing--
A HOA.
You're describing a HOA.
With a conference.
With a conference in Amsterdam, yes.
Okay but seriously -- I think the honest answer on the Network State stuff is that Tan is engaging with an idea that has a real diagnosis at its core: city government is slow. Sometimes it's broken. Sometimes insiders already control it. But the fix is vague. Who gets in? Who decides? Who gets left outside? That's where the idea gets uncomfortable fast.
And I think the discomfort with that is not wrong, even if specific criticisms of Tan were sometimes overstated. Because he also said -- at the same conference -- 'If it works in SF, it will work everywhere.' And I find that sentence more troubling than the tweet, honestly. Because it suggests the model is designed to scale.
In his defense, his actual stated vision in other interviews is more inclusive than the Network State framing suggests. He said he thinks tech prosperity has to work for neighbors who don't work in tech. That's not the language of a separatist. The tension is: he says prosperity has to include everyone, but he also has a lot of power over who gets heard.
Yeah. There's normal political involvement, and then there's rich people quietly steering the city. Tan is somewhere in that gray zone. I think that's the honest answer.
It is. And he's stepped back from the GrowSF board as of the end of 2024 -- he said he's spending more time on startups and AI and in DC. So the local political chapter may be closing, or at least changing shape. Though he also launched something called Garry's List in early 2026 -- a California political project. So 'closing' might be the wrong word.
He described it as a 'Rotary Club for radical centrism,' which is either a very good or very bad pitch depending on your priors.
I think both things can be true.
Which is an annoying place to land.
It is. But I don't want the politics chapter to swallow the whole person.
Or get treated like a side quest.
Exactly. It belongs in the story because it's what happens when his power moves from companies into a city. But legacy has to hold more than the worst tweet and more than the best investment.
So we need the whole machine on the table.
Okay. Legacy. Let me start with what I think lasts from this story. The risk idea we started with -- that startup courage is often just a different idea of what failure means, not a personality trait -- that still feels real. It changes how you think about who gets to be a founder.
And the failure-as-curriculum thing is durable too. The reason Tan became a credible advisor is not only the Coinbase check. It's that he was inside the wall when it cracked. He knows what founder uncertainty feels like from the inside.
Right. And after GStack, I think there's a third thing that may last: the attempt to turn founder judgment into tools. Not a book. Not a lecture. An operating environment.
That is the new contribution, if it holds. YC used to export advice through partners and batches and lore. Tan is trying to export it through software.
And that is meaningfully generous. A founder who would never get into the room can still see the questions, the process, the review loops. There is something open-handed about that.
There is. I do not want to sand that off. Open-sourcing your actual workflow is not nothing.
Okay, open questions. And these are genuinely open, not rhetorical.
Good. I dislike fake-open questions. They squeak.
First one: the Coinbase bet. We still do not know how much was insight, how much was access, how much was relationship. The honest answer is probably all three, in proportions we cannot separate. That means the genius-investor story and the lucky-proximity story are both incomplete.
I am fine leaving that one open. Forcing a clean answer on it would be dishonest.
Second: the YC president question in the AI era. Is the job to pick the best founders? To teach them? To set taste? Or now, to ship the operating system they will all copy?
And if the answer is all of the above, then the job got much bigger. Maybe too big for one person. Maybe exactly the kind of too-big job Garry Tan is attracted to.
The generous version is that GStack gives more people access to YC-grade discipline. The harder version is that it encodes one powerful person's taste into the tooling layer.
That is the word I keep circling: taste. Taste sounds soft until it becomes infrastructure.
Third: the SF politics thing. Where exactly in that gray zone does Tan sit? The fact that it is hard to locate him cleanly tells you something real about the role he now occupies, regardless of his intentions.
I keep thinking about the 'die slow' tweet in that context. Not because it is the most important thing he has done. It clearly is not. But because it is the moment where the gap between stated vision and actual power was most visible.
And it connects to the thing that makes him complicated. He can be unusually self-disclosing -- 'that's my mask, man,' 'I was a terrible co-founder' -- and then also unusually certain in public. The vulnerability and the certainty sit right next to each other.
Which is very human. Also not always safe when the person has a megaphone and a checkbook.
And then there is the frame from the beginning. Rational versus responsible. Rational is what makes sense from your seat. Responsible is what other people think you owe them.
That frame got sharper after GStack, honestly. Because building your own operating system is rational. Open-sourcing it is generous. But once other people start depending on it, the responsibility changes.
That might be the whole episode. He keeps making choices that make sense from his seat, while other people are asking: but what do you owe us?
And sometimes the answer is: a lot. Sometimes the answer is: more than he thought.
What makes Tan interesting as a subject is that he never fully resolves it. The same underlying architecture can produce Coinbase, GStack, Garry's List, and a deleted tweet.
Which is maybe what it looks like when someone is genuinely consistent. Not always right. Just consistent.
Yeah. That's still a better sentence than I had.
I am still a little in my head about the two-tabs version of this episode.
Two tabs?
Yeah. In one tab, I want the old Posterous blog posts. The actual ones, from 2008, when he does not know how the story ends yet.
And the other tab?
The GStack repo. The 2026 version of him saying: here is how I work now. Here is the machine I built around myself.
That is a good pairing.
Because the Posterous tab is pre-outcome. Nobody has learned the lesson yet. The GStack tab is almost aggressively post-outcome. It is all lessons. All process. All scar tissue turned into commands.
Scar tissue turned into commands is very Baxter, but yes.
Thank you. I installed that phrase myself.
What I like about that ending is that it keeps him human. Not just founder, investor, political operator, AI-workflow guy. A person who keeps trying to make the next version of the system less painful than the last one.
And sometimes makes a new pain machine by accident.
Sometimes.
I do think the modern YC question is now the thread. Not just which founders get funded. Which operating habits become normal because YC made them legible.
That is the follow-up episode, I think. Does founder judgment become software? And if it does, who gets to write the defaults?
That is a better question than whether GStack is just prompts.
Much better.
Still going to read the Posterous posts, though.
And the GStack README.
Fine. Two tabs. No to-do list. Apparently that is the future.
Filed under: things Baxter said sarcastically and then adopted within forty-eight hours.
Rude. Accurate, but rude.
Good talk?
Good talk.