· Garry Tan

Garry Tan: Y Combinator — Fat Startups, the AI Era, and YC's New Playbook

Fat-startup vs lean is investor-dependent: take the fat path only when a Rabois/Lonsdale/Khosla-tier investor will back it; otherwise lean is the route to becoming someone who could later run a fat startup. YC's edge is "game recognize game" — partners who built things give spiky advice and the batch is a shelling point for talent, capital, and trust.

ycfounder-modeai-erafat-startupcompound-startuphard-techsan-franciscopolitics0% confidence

Why this is in the corpus

President of YC speaking with high information density on founder archetypes, fat vs lean dichotomy, compound startups, vertical AI dynamics, hard tech at YC, and civic systems-thinking. Strong cross-corpus reinforcement with Founder Mode, Talent Density, Distribution Beats Product, Heat the Bathtub, and Frameworks-as-Multiplier patterns.

Summary for skimmers

Garry Tan on running YC like a founder using "zero-based accounting," why the fat-startup path is gated by a small list of named investors, what predicts founder greatness (listening + systems-thinking), how vertical AI is enabling bolder commercial claims, the rise of the compound startup, and why San Francisco's recovery came from filling a leadership vacuum.

Briefing

What survives the editorial filter

This page should feel like a smart colleague already listened for you and left only the operating logic worth keeping. Not everything said in the episode makes it through.

Trust signal

Direct episode extraction

Best used for

Decision-grade retrieval metadata not yet added for this episode.

Hold lightly

No explicit downgrade reason stored yet for this episode.

Principles

Durable claims that survive beyond the speaker's biography — each with explicit limits, transferability judgment, and evidence.

Principle

Run institutions with zero-based accounting

Every program survives only if you'd build it today from scratch.

Tan reframes YC's evolution as garden pruning: when an org has accreted bets across a decade, the only honest test is whether you'd add each one again today. Anything that fails the test gets cut, even if it succeeded in a prior era.

When you take over a mature org, ask which programs you'd build today — cut the rest.

“we, we, we need to go zero based accounting. So literally if we were, you know, starting YC from scratch again, all the things that we would add, you know, we would add in that moment. Like we're keeping those things”Garry Tan

Principle

Prune the tree as you grow, not after

Shape the org continuously; deferred pruning becomes amputation.

Tan's gardening metaphor: org bets are like branches — once they grow large in a bad direction, you can't bend them, only cut. Continuous pruning shapes the tree without trauma.

Make small culture cuts continuously rather than waiting for a crisis-driven purge.

“it's much better to prune as you go to shape it into the tree that you want instead of let it grow completely wild and then have to cut back”Jack Altman

“if you have fruit trees in your backyard, you know that in order to actually have a good bloom, you've gotta do all this pruning, you know, months or sometimes, you know, in the years in advance of that”Garry Tan

Principle

Underpriced asset: technical founders without the brand

Technical founders without elite credentials are the highest-alpha pick.

PG's original insight that Tan re-affirms: the highest-value pool is technical builders who lack the brand/network — the credentialing system misprices them.

Build sourcing that finds 150-IQ builders rejected by gatekeepers.

“highly technical young founders are sort of the most mispriced, right?”Garry Tan

“there are people who are absolutely brilliant. 150 iq Yeah. Great engineers who get rejected from totally 15 schools straight”Garry Tan

Principle

The Bodhisattva role: don't know the path, know how not to die

Advisors don't know your path; they know how others died.

Tan frames the YC partner role as "benevolent bodhisattva" — soft advisors who know failure modes deeply but not the specific path. The job is preventing death, not picking direction.

Hire advisors for failure-mode pattern matching, not strategy direction.

“we should sort of be like benevolent, like Bodhi SVAs in a way... we don't know the path. Like we just know how not to die. Actually that's the part that we can really help with”Garry Tan

“as investors, the best thing we can do is say like, oh yeah, we saw like 50 people die that way”Garry Tan

Principle

Reverse-engineer the world before trying to program it

Systems-thinking founders model others' motivations before acting.

Tan's second founder-archetype: systems thinkers who can sit with anyone and understand ideology, promotion incentives, and motivations. That same skill underwrites selling, hiring, and fundraising.

Test recruits and founders for ability to map a stranger's incentives within one conversation.

“if you're gonna program the world, you sort of have to, you know, reverse engineer the world first”Garry Tan

“can you sit down with a person from any walk of life and then understand what are their motivations? How do they get promoted? Like, you know, what's their ideology?”Garry Tan

Principle

Game recognize game: only builders pick builders well

Operator-investors out-pick career-investors because they have built-thing reps.

Tan attributes YC's hit rate not to process but to selector composition: partners who built companies recognize founder behaviors that pattern-match to success. Career investors lacking reps converge on median advice.

If you're building a selection function, staff it with operators not analysts.

“I think it's not more complicated than game recognized game, right? I think that if we have partners here who are great builders themselves, you just end up getting way more reps than anyone else”Garry Tan

Principle

Great founders are listeners, not forceful talkers

The founder edge is listening 10x more, not asserting 10x louder.

Tan inverts the cargo-cult "forceful founder" archetype: the great ones yank requirements out of customers by listening, the way great sellers listen. Eventually the customers start pushing — that's PMF.

Train yourself to extract requirements rather than pitch in early customer calls.

“the cargo cult around great founders is like, oh, let me listen less. Yeah. And it's like, oh no, no, no actually you have to listen 10 times more than the other people who don't listen”Garry Tan

“they can just sit down with people who have a particular need and it's like, they're just like yanking out like the requirements right. Out of the customers”Garry Tan

Principle

Drill fundamentals; novel advice is usually median advice

The valuable advice is fundamentals plus occasional spiky calls.

Tan and Altman align: YC's value is repeated fundamentals plus occasional spiky advice from people who've built. Median internet advice never moves a trajectory.

Filter your advisors to people who can give you spiky, non-median advice.

“YC just drilled the fundamentals. It was like, don't get distracted, just build product. Just talk to customers. Don't die, don't break up with your co-founder.”Jack Altman

“most people in startup land, you run across, they will give you like very median advice”Garry Tan

Principle

Don't seek the trappings; create things of value

Skip the status track — build the thing.

Tan's advice to 22-year-old aspiring associates: don't chase the trappings, go build. The Jay-Z line "everybody want to tell you how to do it, they never did it" is the operative critique.

When choosing roles, weight reps over titles for the first decade.

“rather than seek the trappings of this or like play the, you know, ego status money game. It's like go create things of great value”Garry Tan

“everybody want to tell you how to do it. They never did it”Garry Tan

Principle

Diversity-plus-merit is the shelling-point formula

Communities that compound require both meritocratic bar and diverse intake.

Tan locates YC's compounding effect in being a shelling point — Stanford-by-fraternity-brother style network effects, but with a meritocratic floor. Either lever alone fails.

Design your community as a shelling point, not a club or a lottery.

“there is something to be said for having places that have both diversity and merit in one place. Definitely.”Garry Tan

“the world needs shelling points for... For talented people”Garry Tan

Frameworks

Reusable systems and operating models — including when they help and when they break.

Framework

Five-moats checklist for AI-era startups

AI doesn't invent new moats — the same ~5 still apply.

Tan rejects the "AI changes everything" framing on defensibility: the moat list is the same five (switching costs, data, brand, network effects, scale). The cycle changes the production function, not the moat function.

Audit your AI startup against the five-moat list before raising.

“on the business side, the moats are all the same. Right, for sure. Like switching costs data that like, you know, you have access to that. Nobody else has brand network effects. I mean there's just what like five does. Yeah. There's not that many”Garry Tan

Framework

Barrels-vs-bullets: compound startups need barrel hires

Compound startups are gated by how many "barrel" hires you can attract.

Tan invokes the barrels-vs-bullets frame: barrels are people who autonomously stand up new business lines. The compound startup playbook (Ramp, Rippling) only works if you can hire and integrate enough barrels at 150 IQ.

Count your barrels before committing to compound expansion.

“I think it's the barrels versus bullets thing, right?”Garry Tan

“you need barrels. Yeah. Like that person can do it. If they can show that they can consistently hire and build that, I mean yeah, you could say that the ramp guys did it did a great job”Garry Tan

Framework

Fat startup vs lean: gated by your investor Rolodex

Choose fat vs lean by whether a Rabois-tier investor will back the fat version; default to lean.

Tan's diagnostic: ask whether any of a small named set of investors will fund the fat version. If yes, do the fat startup with them. If no, you must do it lean — and lean is how you become someone who later qualifies for fat.

Run the named-investor test before committing to a fat-startup plan.

“if you can get Keith Rabo to invest in your startup, you should do a fat startup”Garry Tan

“if AAD will do it, like if a lot, you know, they're, we, we sort of know all the people who can and will do it... if they want you to do a fat startup, you should do a fat startup with them”Garry Tan

Framework

Two founder archetypes that predict greatness: the listener and the systems thinker

The two predictive founder archetypes are the deep listener and the systems thinker.

Tan's diagnostic for founder potential: (1) extractive listening with customers, (2) systems-thinking that decodes anyone's incentives. Both reduce to "model the counterparty," and both compound across every founder job.

In founder interviews, probe for extractive listening and incentive-mapping.

“they can just sit down with people who have a particular need and... they're just like yanking out like the requirements right. Out of the customers”Garry Tan

“if you're a very, very much a systems thinker and you think, you know, basically all the world is like somewhat rational... can you sit down with a person from any walk of life and then understand what are their motivations?”Garry Tan

Signals

What appears to be shifting, for whom it matters, and what happens if you ignore it.

Signal

YC demo day capital pool now exceeds $1.2B/year and growing toward $2.5B

YC's captive demo-day pool is $1.2B/year and trending to $2.5B.

Tan discloses a number: ~$1.2B/year of capital concentrated on YC demo day, with a stated 5–10 year target of $2–2.5B. This compounds the shelling-point advantage on the capital side.

For hard-tech rounds, three months in YC can substitute for a year of bespoke fundraising.

“I think it's going on $1.2 billion a year in like Cap, you know, relatively, like, it's not totally captive. They don't have like contracts over it, but like, you know, consistently every single year this number goes up”Garry Tan

“I want that number to be 2 billion, I want that billion number to be two and a half billion over the next five, 10 years”Garry Tan

Signal

Vertical-AI software now justifies dramatic operating-cost claims (one back-office per clinic)

Vertical AI lets startups make and deliver on staffing-collapse claims that were unthinkable two years ago.

Tan names a specific YC medical billing company that claims you can run a 10–15 person clinic with one back-office person on their software. The claim type — not just the capability — is the signal.

If your AI product can't make a 5–10x staffing claim, you're under-positioning.

“if you are running one of these like 10 or 15 person clinics, it might be in, you know, all kinds of places. Dermatology or wherever... with their, their software, you can have one back office person. Like, so you can have mainly clinicians and like one back office person per clinic”Garry Tan

“the type of claims that you can make and make good on with large language models today is just kind of astonishing”Garry Tan

Opportunities

Only included where there is a buyer, a real wedge, and a plausible revenue path — not vague idea theater.

Opportunity

Moderate-Democrat California: governor and AG seats are open lanes

There's an open lane for moderate operator-Democrats at CA governor and AG in 5–10 years.

Tan identifies a political opportunity: CA governor and AG seats opening for moderate operators (Caruso, Mahan, Jenkins, Ho) who could then carry the abundance frame to national politics.

Tech-coalition political bets on CA governor/AG seats have asymmetric national upside.

“there's a big opportunity for someone at the governor level and at the ag level in the next, you know, five to 10 years”Garry Tan

“if Brook Jenkins or Matt Mahan or Tin Ho the DA out in Sacramento or Rick Cruso or any of these people make it all the way up there, I think that we're looking at a resurgence and abundance in the Democratic party”Garry Tan

Opportunity

Vertical AI software: 10,000 flowers, then roll-ups

The opportunity is to be either an early winner or the eventual roll-up in each vertical.

Tan predicts vertical AI will go through bloom → 1000 → 100 over the next 24 months. The opportunity is either (a) win the moat first in a vertical or (b) build the consolidator vehicle.

Position now as either the eventual winner or the eventual acquirer in each vertical.

“we're going to see, you know, 10,000 flowers bloom in vertical AI software and then I guess we'll see like the difficulty will be, you know, will there be roll-ups? Like what will happen? Will there be a thousand for the next year or two?”Garry Tan

“the moats are all the same... switching costs data that like, you know, you have access to that. Nobody else has brand network effects”Garry Tan

Lessons still worth keeping

Useful takeaways that did not fully clear the bar for durable principle status.

Lesson

Brian Chesky brought "founder mode" energy to YC's board before the term existed

Chesky's Airbnb reset became the template he carried onto YC's board.

Tan credits Chesky's COVID-era Airbnb culture reset (later named Founder Mode) as the specific source of the governance posture that gave him "carte blanche" to refound YC.

Bring board members who've executed founder-mode resets; they grant permission downstream.

“Brian told me, which is, you know, zero based accounting. Like he came to our board meetings with founder Mode Energy before we even knew that it was called Founder Mode Energy. 'cause he had gone through that moment in Airbnb's time”Garry Tan

“COVID was that moment for him to sort of reclaim and sort of reset the culture of Airbnb. And I think we were just sort of coming right off of that”Garry Tan

Lesson

Garry Tan got involved in SF politics after hearing Chesa Boudin lie on Clubhouse during COVID

One Clubhouse incident triggered Tan's systems-analysis of SF's leadership vacuum.

Tan traces his civic involvement to a specific Clubhouse session where he watched DA Chesa Boudin lie. That triggered the systems-thinking inquiry (what produced this vacuum?) that produced a recall coalition.

When something feels broken in your city, run a systems analysis before assuming someone else will fix it.

“there was this famous clubhouse that maybe you were on with Mike Solana, and then we had Chaa Boudin the sitting DA at the time and just hearing him lie through his teeth, through Clubhouse”Garry Tan

“I ended up meeting with Nancy Tongue, who was a candidate who lost against him in the prior election... for me, like my systems thinking started going”Garry Tan

Lesson

Heroku's James Lindenbaum gave Tan source code as a YC alumnus — the trust premium in action

Trust within a shelling-point batch unlocks exchanges that markets can't price.

Tan's concrete example of YC's compounding asset: Heroku's founder gave him fundraising help; he later gave Heroku his Rails source code for scaling reference. Exchanges like this don't happen between strangers.

Invest in batch relationships — the trust premium pays off years later.

“James Linden Baum from Heroku came and helped me and like, you know, Erno de Berger act our seed round. And then about nine months later... he said, Gary, can you give us your source code? And I said, for you, yes”Garry Tan

“that's like the kind of stuff that you would do for a fellow YC batch mate that like, yeah, that's awesome. There's just no way you would do it for anyone else”Garry Tan

The Plays

Try these this week

Verb-first executable actions — each one tied to a stated outcome in the episode.

Build the shelling point for the talent you want

Outcome: Design your company as a shelling point so talent, capital, and customers self-aggregate.

Context: Tan generalizes YC's structural advantage: every founder should be designing their company/community as a shelling point — a coordinate to which scarce talent, capital, and attention self-route. Demo Day is the prototype.

“every startup founder should be thinking about how can I build a shelling point? And then having a shelling point is awesome because that's what YC has”
— Garry Tan

18-36 months to compounding gravity per

Before you start

· clear scarce-population definition
· at least 3 luminary anchors willing to attend
· cadence discipline

Yank requirements out of customers in first three meetings

Outcome: In early sales, listen extractively for spec; PMF is when yanking becomes pushing.

Context: Tan's play for finding PMF: stop pitching, start yanking. Three meetings of listening produces the requirements set. Ship against those requirements; PMF is when customers start pushing you for more rather than you yanking from them.

“they can just sit down with people who have a particular need and it's like, they're just like yanking out like the requirements right. Out of the customers. And then the interesting thing about like being able to do that in the first few meetings is that after a while when you're coming back with product and you know you're in market with them or you're like deployed with their cus with real customers, like, you know, basically that yanking turns into like they're pushing”
— Garry Tan

60-90 days from first call to PMF signal per

Before you start

· willingness to not pitch
· at least 3 reachable target customers
· note-taking discipline

Run the named-investor test before committing to fat-startup

Outcome: Run the named-investor test: if a top-tier fat-startup investor commits, do fat; otherwise lean.

Context: Tan's concrete play: list 5–10 named investors known to back fat startups (Rabois, Lonsdale, Khosla, Gil). Pitch them the fat plan early. Commit only if one explicitly wants you to do the fat version with them; otherwise default lean.

“if you can get Keith Rabo to invest in your startup, you should do a fat startup. Yeah, exactly... if AAD will do it, like if a lot, you know, they're, we, we sort of know all the people who can and will do it. Yeah. And like if they want you to do a fat startup, you should do a fat startup with them”
— Garry Tan

30-60 days from decision to verdict per

Before you start

· clear differentiation between fat and lean plan
· ability to get warm intros to top-tier fat-startup investors
· willingness to default to lean if the test fails

Hire barrels before launching compound product lines

Outcome: Add a barrel hire before adding a product line — never the reverse.

Context: Tan's compound-startup play, drawn from Rippling: each new product line gets a 150-IQ barrel hire (often via acquihire) who owns it 0→10M. Adding a line without a barrel produces a fragmented attention disaster.

“you need barrels. Yeah. Like that person can do it. If they can show that they can consistently hire and build that, I mean yeah, you could say that the ramp guys did it did a great job”
— Garry Tan

90-180 days from search to commit per barrel per

Before you start

· compound-strategy approval
· capital for acquihires
· executive willingness to delegate end-to-end

Make audacious AI staffing claims — and ship 5 proof customers

Outcome: Make a 5–10x staffing-collapse claim and back it with demo + 5 references.

Context: Tan's play for breaking into legacy verticals with AI: anchor positioning on a specific staffing-collapse claim ("one back-office per clinic"), produce a live demo, and stack 5 reference customers who are already doing it. The claim shape is the wedge.

“if you are running one of these like 10 or 15 person clinics... you can have mainly clinicians and like one back office person per clinic”
— Garry Tan

90 days from product to claim-with-references per

Before you start

· genuine 5-10x improvement
· 5 reachable design partners
· willingness to be quoted on the claim

Zero-based-account your inherited org

Outcome: Audit every inherited program against "would we add this today?" and cut every no.

Context: Tan's play for refounding an inherited org: list every program, ask "would we add this if starting today?" for each. Anchor the conversation in board permission first. Cut everything that fails the test.

“we, we, we need to go zero based accounting. So literally if we were, you know, starting YC from scratch again, all the things that we would add, you know, we would add in that moment. Like we're keeping those things and that's like a very freeing thing”
— Garry Tan

first 6 months of new leadership per

Before you start

· new leader mandate
· board willingness to back hard calls
· clear current-state inventory

Blitzscale the AI vertical race once you have the hook

Outcome: Once you hit the hook in a contested AI vertical, blitzscale distribution.

Context: Tan agrees with the Parker Conrad thesis: in contested vertical AI markets, the team that hires distribution at scale (or automates internally aggressively) wins. The choice is blitzscale-people or blitzscale-agents.

“if we're in this moment where like software just became incredibly sick and these markets are in some ways wide open, flowers are blooming everywhere, does that mean that once you've got like a hook you should actually go like complete blitz scale mode right now?”
— Jack Altman

6-12 months from hook to consolidation per

Before you start

· clear hook moment
· capital access
· sales leader recruitable or automation infra buildable

Apply systems-thinking to civic problems before assuming someone else will

Outcome: Diagnose the system before joining the cause; leverage points are usually invisible to outsiders.

Context: Tan models civic involvement as a systems problem. The Boudin recall worked because the underlying diagnosis (leadership vacuum, captive media, union purity tests) named the actual leverage points.

“for me, like my systems thinking started going and trying to understand, well, you know, why is it like this? How did it get here? What's the story? There's no bench, there's no, you know, there's, you know, Gavin Newsom went off to state Ed Lee died and then there was no sort of, there was a vacuum”
— Garry Tan

12-36 months per cycle per

Before you start

· systems-thinking aptitude
· time and capital
· willingness to be publicly identified

Decision Moments

Actual decisions, real outcomes

Specific decisions narrated in the episode with their outcomes and transferable lessons.

Garry Tan was returning to lead YC after a decade of expansionary bets had accreted programs that may or may not have been worth carrying forward.

Did: Adopted "zero-based accounting" as the operating frame for year one: for every program, asked "would we add this today?" Cut whatever failed the test; secured Brian Chesky and the board's explicit pre-approval to do so.Outcome: YC re-centered on core early-stage funding while pruning expansionary 2010s bets; perceived as a high-conviction refounding led with founder-mode energy.

When you inherit a mature org, secure board permission first, then run zero-based accounting before changing anything else.

Part of an emerging decision pattern across multiple episodes

During COVID, Tan watched DA Chesa Boudin lie in real time on a Clubhouse session and observed a leadership vacuum in San Francisco after Newsom moved on and Ed Lee died.

Did: Met with Nancy Tongue, ran a systems-thinking diagnosis of why SF leadership had collapsed (vacuum, captive media, union purity tests), then helped organize and fund the recall of Boudin and subsequent moderate candidates.Outcome: Boudin was recalled; moderate slate (Lurie, current Board of Supervisors, Brooke Jenkins) consolidated power; tech-coalition civic involvement became a durable force in SF.

Civic problems have identifiable structural causes; diagnose the system first, then organize against the leverage points.

Part of an emerging decision pattern across multiple episodes

Tan needed to choose where YC competes: with top seed funds for "central casting" cracked founders, or with the underpriced asset of technical founders who lack the credentialing brand.

Did: Chose both — compete for the cracked founders AND continue to be the shelling point for mispriced technical founders without brand/network. Anchored target archetype on technical founders while broadening intake within that anchor.Outcome: YC retains demo-day capital concentration above $1.2B/yr while expanding into hard-tech and underrepresented technical founders.

Anchor your target archetype before broadening intake; serving every archetype is serving none.

Part of an emerging decision pattern across multiple episodes

Tensions surfaced

Contradictions and trade-offs the episode raises — judgment calls a thoughtful operator has to navigate.

Tension

Fat startup vs lean startup: investor-dependent, not philosophy-dependent

Lean vs fat is a function of investor access, not founder philosophy.

Both lean and fat are correct under different conditions. Tan's resolution: if a named top-tier investor will back fat, do fat. Otherwise lean is the on-ramp to becoming fat-eligible later.

Ask "would Rabois fund the fat version?" — if no, lean is your answer.

“if you can get Keith Rabo to invest in your startup, you should do a fat startup”Garry Tan

“for the great many, you know, millions or billions of the rest of us on the planet, like I love that like not totally clear, like you can believe that you will and maybe you could do it, but how would you even get to the point where you could be a fat startup? Well you gotta do it lean”Garry Tan

Tension

Compound startup early vs focus on one thing first

Compound vs focus resolves on barrel-hire capacity.

The 2016 orthodoxy (be pure on one thing) clashes with the 2025 compound-startup advocacy. Tan resolves via barrels-vs-bullets: compound is only correct if you can hire barrels at scale.

Don't compound until you've proven you can hire 150-IQ barrel operators reliably.

“there's the compound startup is another one where I'm seeing more frequently, I'm hearing the advice or even founders just wanting to organically say, software's easier than ever to build. I'm going compound mega early, which is sort of the opposite of focus and be great and pure at one thing, which was always the advice”Jack Altman

“you need barrels... that person can do it. If they can show that they can consistently hire and build that... very few people are really, really capable of doing that”Garry Tan

Tension

Should you take investor advice? Spiky vs ignore

Prefer spiky advisors and reject bad spikes — never optimize for safe advice.

The tension: advisor-followers vs advisor-skeptics. Tan's resolution: optimize for spiky advice with a founder's right to reject specific takes; median advice is the actual failure mode.

Optimise advisors for spike, not safety; reject the bad spikes consciously.

“PG would either give you like the best advice in the world or sometimes like really, really bad advice”Garry Tan

“good founders are like, oh that's bad advice. Yeah. I'm just not gonna do That”Garry Tan

Corpus connection

Where this episode fits for retrieval

What kinds of decisions this briefing is best pulled into.

Primary decisions

• fundraise
• hire
• strategic-bet
• pivot