Keith Rabois
The month in one sentence
Rabois spent the month arguing that judgment beats systems, winners are unusually obsessive, institutions frequently make technology worse through bad implementation, and America should treat Chinese AI as a strategic adversary rather than merely another competitor—with a large side helping of soccer rage, venture-capital inside baseball, and deliberately abrasive one-liners.
What makes the feed interesting is that these subjects are not actually disconnected. His worldview is remarkably consistent across startups, sports, politics, and AI: exceptional outcomes come from exceptional individuals exercising judgment; bureaucracy and proceduralism tend to destroy that judgment; incentives matter more than stated principles; and competition is real enough that sometimes you should stop pretending everyone is playing a friendly game.
1. His most substantive startup idea: hire a near-peer, not another VP
The best business discussion of the month comes from Gokul Rajaram arguing that a founder around $50–300M ARR should hire a “near-peer”: someone good enough that the founder could plausibly report to them, complementary to the founder, capable of operating at the next scale, and able to absorb the coordination burden that otherwise turns the founder into the bottleneck.
Rabois strongly endorses it: “Excellent advice.” More interestingly, he says he had just presented on the same topic at a company offsite and keeps probing the idea:
How does a founder identify the right fit? Are there good examples from the last three years? * How should founders actually source candidates?
Rajaram's follow-up gives a very Rabois-compatible description of the ideal executive: someone who can jump from strategy to microscopic execution, understand both product and business, and remain a genuine doer rather than becoming a capital-E “Executive.” Rabois explicitly agrees with that first characteristic.
This is interesting because it exposes what Rabois appears to think management actually is. He doesn't seem to believe that scaling means becoming more abstract. The opposite: the best senior people must be able to traverse abstraction levels better than everyone else.
That connects cleanly to his old operator reputation. The executive he admires is not merely a coordinator of specialists. They can discuss strategy at the board meeting and then descend five layers into why a particular project is failing.
2. Venture capital: judgment remains the scarce thing
Several short exchanges point to the same investing philosophy.
Someone argues that AI and data will not eliminate venture alpha and cites Google Ventures as a supposedly data-driven investor. Rabois responds that their best investments “all didn’t use any data.”
Josh Wolfe says that today's genuinely exceptional founders can justify extraordinary valuations, but that they are exceptions and that 99% of companies funded at today's prices will erase investor capital. Rabois simply replies: “true.”
When Pavel Prata summarizes observations from VCs and LPs—too many emerging managers constructing concentrated portfolios without justification, seed investing remaining extremely power-law distributed, funds growing larger without reason, apprenticeship lineage mattering, and former builders outperforming institutional allocators—Rabois calls them “sharp insights.”
And when Helen Min says New York investors are hearing LP pressure to spend more time in San Francisco because of AI, Rabois replies:
“don’t invest based upon LP advice, ever.”
The common theme is almost anti-financialization. Venture investing, in his conception, is not primarily an optimization problem. It is a judgment problem under extreme uncertainty, and importing more formulas, committees, portfolio conventions, or LP opinions can make you worse rather than better.
Even his reply to Paul Graham fits. Graham celebrates how SAFEs have scaled from tiny startup rounds to hundred-million-dollar financings. Rabois says, essentially, that doesn't mean using them that way is wise: “that isn’t a good idea. Ask Jonathan.”
So his instinct is consistently: a mechanism succeeding at one scale does not imply that extending it everywhere is intelligent.
3. Ramp remains his canonical example of startup excellence
Ramp appears several times.
He quotes David Senra on “The magic of Ramp.” He posts Ramp's first board deck exactly seven years after the first meeting. At month's end, someone praises Ramp for retaining early employees and developing them into executives, and Rabois replies simply:
“This.”
That final point is more revealing than it looks. It complements the near-peer discussion but introduces a second path: great companies don't merely recruit polished executives from elsewhere; they compound their own people.
The implicit model looks something like:
exceptional founder → exceptional early employees → enormous responsibility → internal development → future executives
rather than:
founder → company grows → replace amateurs with professional managers.
This is one of the recurring Rabois ideas worth noticing because it runs against the default “hire the experienced person who has already done the job” playbook—even though he simultaneously believes that at sufficiently large scale the founder may need one extremely experienced near-peer.
Those positions aren't contradictory. His apparent prescription is: grow most leaders internally, but make the very small number of external executive hires unusually consequential.
4. “No Days Off” is not motivational rhetoric to him
One of the funniest and most revealing episodes begins when Rabois quotes a story that Lionel Messi arrives at training around 6:50 a.m., three hours early:
“No days off.”
Someone pushes back: elite athletes do take recovery periods; business people misunderstand athletic training if they think greatness means going 100% for 365 days.
Rabois goes ballistic.
He calls that claim false, invokes Kobe Bryant, tells people to study the derivation of “No Days Off,” brings up Michael Jordan shooting at 6 a.m. after a playoff game, recommends The Jordan Rules, Hard Drive, and Tim Grover's Winning, cites having trained around elite athletes himself, and eventually responds to criticism with:
“blah blah blah. Loser.”
This is partly comedy—the sheer escalation is disproportionate—but it also makes one of his deepest beliefs unusually explicit.
He doesn't interpret extraordinary work habits as survivorship-bias anecdotes attached to already-gifted people. He appears to think the obsession is causally central to becoming a winner.
His model is not “work intelligently, recover, maintain balance, and maximize sustainable long-run output.” It is closer to:
the people who become historically great have a qualitatively different relationship with work from normal people, and attempts to normalize their behavior misunderstand the phenomenon.
Whether that is empirically right in every domain is separate. What is interesting is how strongly he believes it—and how readily he carries the same model between Jordan, Kobe, Messi, founders, and companies.
5. Soccer becomes a case study in how technology can make a system worse
Rabois spends an enormous amount of the month arguing about VAR and officiating.
His initial claim is surprisingly general:
Technology has mostly improved baseball, football, hockey, and tennis, while making the NBA and soccer materially worse.
People naturally respond that the technology itself isn't the problem; implementation is.
Rabois's answer:
“those are the same.”
That may be the most interesting line in the entire sports discussion.
His broader claim is that evaluating a technology separately from the institution implementing it is meaningless. Someone tells him that VAR can work well in different leagues and rugby's TMO is well-liked, so the underlying technology shouldn't be blamed.
Rabois responds that this is exactly his macro point about technology “since the beginning of time.”
In other words, technology isn't an exogenous magic substance that raises performance. It gets embedded into organizations with incentives, rules, human operators, latency, edge cases, and authority structures. Sometimes providing officials with more information and more intervention opportunities makes the product worse and perhaps even makes decisions less accurate.
He repeatedly says video replay has magnified bad NBA officiating rather than fixed it. In soccer, he complains that VAR ruins flow and creates interventions that would never have occurred under ordinary refereeing.
The interesting broader analogy is obvious: this is also how he appears to think about AI, regulation, management systems, and quantitative venture investing. More information or more technology does not automatically produce better judgment.
6. He becomes completely consumed by the World Cup
For several days, the account almost turns into a US soccer message board.
He disputes a red card, argues that VAR procedure was improperly followed, says the assistant shouldn't have buzzed the referee and that video should have been shown at normal speed, and insists contact was initiated against the American player rather than constituting a foul.
When someone says the decision was legitimate because VAR reviews red cards, Rabois repeatedly insists that there wasn't originally a red card to review.
Then the US loses and he turns on the goalkeeper.
A supporter says the coach couldn't be blamed for a goalkeeper clearance. Rabois says the coach should have played the other keeper. When told no goalkeeper would have saved the team, he says perhaps not, but two and maybe all three goals were saveable—and says he actually drafted a halftime tweet calling for the goalkeeper to be pulled but didn't post it.
Later:
“Watch the final 8 seconds. Last night was not an aberration.”
When someone responds that the match had been lost long before that:
“yes but he should not have been in net.”
This is classic sports-fan tunnel vision, but it is also very Rabois: once he thinks he has identified the decisive personnel error, he keeps hammering it.
There are good throwaway jokes too. Delian Asparouhov observes that every team winning a World Cup match has scored more goals than its opponent. Rabois replies:
“you are now ready to be a growth investor:)”
And when David Marcus says Americans don't truly care about soccer because an awful US performance didn't produce boos, Rabois answers:
“Knicks fans def would have.”
7. His theory of sports is really a theory of agency
Another soccer argument concerns whether a spectacular pass was skill or luck. Someone says people overinterpret a successful cross because many intended crosses fail.
Rabois insists:
“he placed the ball in the spot that only his teammate could play.”
This fits perfectly with another quote he amplifies: a clip presented as evidence against the idea that winning is “luck.”
Across investing and athletics, he has a very strong agency prior. When exceptional people repeatedly outperform, he wants to explain the outcome through skill, intention, preparation, selection, and judgment—not randomness.
That is intellectually interesting because venture capital and soccer are both domains with enormous stochastic components. Rabois seems almost temperamentally hostile to explanations that assign too much causal weight to luck.
You can see the virtue and danger simultaneously:
It forces you to look for hidden skill where others lazily say “luck.” It can also make you underestimate genuine variance and retrospective storytelling.
That tension runs underneath much of his thinking.
8. AI and China becomes the month's dominant serious argument
The second half of the month is increasingly about Chinese AI models, open weights, distillation, intellectual property, and national security.
Rabois's position is more nuanced than “ban open source,” although he is considerably more restrictionist than many Silicon Valley participants.
He endorses Stratechery's framing that Chinese models and open-source AI are different questions. Martin Casado similarly says people are conflating two issues: open source is foundational to a healthy industry, whereas Chinese subsidization or dumping can occur regardless of whether models are open. Rabois replies:
“correct.”
So the argument isn't:
open models = bad.
It is:
foreign strategic competitors + open models = a special geopolitical problem that ordinary open-source intuitions may not solve.
That distinction matters.
9. He thinks Chinese models may constitute an asymmetric strategic weapon
One argument he endorses says defaults matter because nearly everyone will use base models rather than fine-tunes, meaning a Chinese base model could become a foreign instrument influencing Americans while operating a growing portion of the economy.
Rabois replies:
“correct.”
He also endorses Jack Altman's framing that people in the open-source-AI debate are often reasoning from completely different premises:
If you expect godlike recursive self-improvement, the question becomes who can safely possess it. If you don't, the debate is primarily about economics, competition, and where profits accrue.
Again Rabois says:
“correct.”
This is useful because it reveals that much of what looks like policy disagreement is, in his view, actually disagreement about the future technological state of the world.
If AI becomes merely very useful software, openness has one risk/reward profile. If AI becomes strategically dominant intelligence infrastructure, the national-security implications are completely different.
10. But his strongest argument is not “national security”; it's rule of law
Someone mocks American frontier labs complaining that Chinese companies distilled their models: normal people won't sympathize with supposedly all-powerful AI companies asking others to protect their homework.
Rabois responds:
“the rule of law is foundational to American exceptionalism.”
That is significant.
He is not willing to say that because Anthropic or another lab is rich, disliked, ideological, or technically sophisticated, alleged theft ceases to matter.
When Parker Conrad argues that politically connected growth investors are trying to protect Anthropic investments by banning Chinese open-source models despite no proven crime, Rabois gives a much more procedural answer than his normal rhetoric would suggest:
They should have to prove the case in court. But if they prove it and Chinese defendants evade jurisdiction, enforcement by the government can be legitimate.
He subsequently says:
IP theft is already a recognized legal concept. Chinese defendants escaping jurisdiction creates an enforcement problem. Evidence should actually establish copied code. Anthropic should release its evidence publicly. * Frontier labs can simply deny API access to customers they believe are distilling them.
So there is a reasonably coherent legal principle underneath the rhetoric:
prove wrongdoing, don't invent a new ad hoc category merely because AI is involved, but don't allow foreign jurisdictional evasion to make American law meaningless.
11. He's considerably more hawkish than most of tech Twitter
Even while distinguishing open source from China, Rabois repeatedly moves toward restriction.
When someone says defensive measures against foreign open-weight models will ultimately be porous and America should instead accelerate compute, energy, talent, infrastructure, and model development, Rabois answers:
“slowing down helps play for time as this positive stuff kicks in.”
Then:
“it has been more successful than people realize (via export controls on hardware).”
When Preston Byrne says there isn't much the US can do to stop Americans using Chinese open-weight models because China isn't subject to reciprocal regulatory enforcement, Rabois says:
“lots of ways to prevent this.”
And by month's end, when advanced foreign robotics imports are discussed, Rabois thinks banning them is a “no-brainer.”
He then says robots are obviously bannable while models are more complicated, but adds that he is probably closer to banning the models too, while believing there are quicker intermediate measures.
So his directional position is clear: economic openness is subordinate to winning a strategic competition with China when he believes the underlying technology has national-security consequences.
12. He doesn't buy the “it's just Anthropic investors protecting their bags” explanation
A recurring counterargument is that Silicon Valley investors only favor restrictions because they own American AI companies.
Rabois rejects this aggressively.
He says the people involved often:
don't like Anthropic as a company, dislike its ideology, despise its CEO, have negligible political contributions, * and are motivated instead by beating China and protecting American companies against foreign anti-competitive behavior.
When challenged, he says he's basing this on “actual knowledge.”
Whether one accepts his inside information or not, the interesting part is the causal claim: he thinks critics are applying an overly simple economic-interest model to people whose motivations he believes are genuinely geopolitical.
At the same time, he can't resist pointing out the reverse incentive problem. Late in the month he says VCs without investments in good models are especially fond of complaining.
So he thinks incentives matter; he just thinks the incentive story being told about his own side is wrong.
13. His AI stance contains a genuine tension
There is a real unresolved tension in these posts.
On one hand:
open source is valuable; Chinese AI and open source should not be conflated; violations should require proof; Anthropic should publish evidence; * ordinary legal process matters.
On the other:
Chinese models may be propaganda infrastructure; slowing China down is strategically useful; hardware export controls work; there are “lots of ways” to prevent Chinese models being used; * outright bans may ultimately be justified.
The interesting question his tweets never fully resolve is what happens when national-security precaution outruns provable IP violations.
His rule-of-law argument is strongest when there is demonstrable theft. His strategic argument is strongest when there isn't necessarily any illegal behavior at all—just a dangerous foreign capability.
Those are different justifications for state action, and the month shows him moving between them without fully constructing the bridge.
14. Politics: confiscation, Brexit, Republicans, and Miami
The political posts are less developed intellectually but highly consistent.
He quotes an attack on proposals for wealth taxes that supposedly begin with billionaires and quickly reach people worth $50 million, adding:
“As predicted….”
The implied slippery-slope claim is straightforward: once wealth taxation becomes politically legitimate, the threshold will fall.
During a dispute involving Belgium, he jokes that listening to Belgian complaints makes the Brexit vote look rational because of “petty EU tyrants in Belgium.”
When someone asks how a politician could possibly have arrived at a bad idea while surrounded by staff and consultants, Rabois answers:
“Republicans:)”
And when someone mocks his old Miami boosterism—“I thought Miami was going to be the VC capital of the world by now?”—he shoots back:
“everyone who is best in any field is there now:)”
It is not really an argument. It's Rabois refusing the premise and escalating the claim for comic effect.
15. A few tiny replies reveal a lot about how he thinks
Some of the month's shortest tweets are unusually diagnostic.
On a founder saying investors advised the company to sell during a bad period:
“some investors.”
Translation: don't implicate him in bad advice.
On someone asking whether he has abandoned Opendoor because he doesn't tweet about it:
“don’t be so stupid.”
On criticism of an old Opendoor valuation prediction:
“looks accurate to me.”
On Google Ventures' supposed data advantage:
“their best investments all didn’t use any data.”
On whether founders should follow LP advice:
“don’t invest based upon LP advice, ever.”
On an argument distinguishing Chinese AI from open source:
“correct.”
His style compresses complicated questions into extremely strong binary judgments. Sometimes that's clarifying; sometimes it leaves most of the argument unstated.
16. The jokes are mostly status jokes
Rabois is funny in a very specific way: he likes reducing an elaborate premise to a brutal status judgment.
The best examples:
The observation that winning soccer teams have scored more goals:
“you are now ready to be a growth investor:)”
A critic of “No Days Off”:
“blah blah blah. Loser.”
A typo-filled escalation from banning foreign robots:
“band the brains too.”
Someone confused by his phrase “Fortune 2000”:
“Russell 2000.”
That last one is effectively an acknowledgment that he'd blended “Fortune 500” and “Russell 2000,” delivered without actually saying “I made a mistake.”
And the Miami criticism:
“everyone who is best in any field is there now:)”
The joke structure is frequently: you give me nuance; I reply with an absurdly confident simplification.
17. What ties the whole month together
The strongest through-line is anti-proceduralism combined with extreme respect for judgment.
Rabois repeatedly prefers:
the referee's original perception over a VAR bureaucracy that manufactures an intervention; a founder or elite operator's judgment over organizational process; venture intuition over data-driven investment machinery; a founder's judgment over LP advice; exceptional individual discipline over generalized work-life prescriptions; specific legal evidence over abstract claims that “distillation feels unfair”; * geopolitical realism over universalist assumptions that technology should simply flow everywhere.
But he isn't purely anti-institutional. The China debate shows the opposite: he wants courts, enforceable judgments, export controls, and possibly government bans when institutions are enforcing rules he believes are substantively correct.
So the more exact principle is:
He dislikes process substituting for judgment, not power itself.
He is perfectly comfortable with enormous institutional power when he believes competent people are applying it toward the right objective.
18. Why this month is interesting
The feed is interesting less because any single claim is novel than because you can see a coherent philosophy being stress-tested against wildly different subjects.
In startups: rare people and qualitative judgment dominate.
In venture: power laws make average-case reasoning dangerous.
In management: elite executives must move effortlessly between abstraction and detail.
In sports: great outcomes come from obsessive preparation and individual agency, not merely luck.
In technology: adding tooling can make systems worse if the institution operating it is incompetent.
In geopolitics: free-market or open-source priors should yield when the counterparty is a strategic adversary.
In law: wealth or unpopularity shouldn't erase ordinary protections against theft.
And across all of them, Rabois has a deep suspicion of explanations that dissolve responsibility into systems, randomness, committees, incentives, or complexity. He continually wants to know:
Who made the decision? Were they good? Did they win? Could someone better have done otherwise?
That is probably the most useful way to read the month. The soccer obsession, the “No Days Off” fight, the near-peer discussion, the LP advice, and the China-AI debate are all variations of the same argument about agency, excellence, judgment, and competition.