Keith Rabois
The month in one sentence
Rabois spent May performing the same role he has played in Silicon Valley for years: extremely high-conviction investor-operator, allergic to hedging, convinced that exceptional outcomes come from exceptional people working exceptionally hard, and perfectly happy to convert disagreements about companies, politics, economics, or war into public bets about who understands reality better.
The substance ranges from AI and venture capital to Miami governance, inflation, Iran, Opendoor, seven-day workweeks, Pierre Lamond, basketball, and law-firm analogies. But the underlying worldview is remarkably coherent.
The most interesting thing is not any individual prediction. It is the decision-making philosophy connecting them.
Conviction as an operating system
The dominant motif is Rabois saying, in one form or another:
I made the call. I still believe the call. The evidence will vindicate me.
His prediction that the Iran war was effectively over becomes a recurring running gag because people keep returning to challenge him. Someone says the war isn't over; Rabois answers, “it is over.” Asked what was won: “everything.” Told that his “war over in 4 days” prediction aged badly: “which was accurate.” Later he corrects someone saying day four: “was day 3 actually.”
This isn't really argument by the end. It's closer to victory-lap trolling.
The same style appears elsewhere:
Someone claims he would have lost an inflation bet because inflation is 3.8%. Rabois says to check Truflation and adds that his bets are “likely brilliant.” Asked about people he has wagered with: “several people:). easy win.” Someone imagines Democrats winning the House and Rabois losing another bet: “no chance. everyone realizes that i am correct now.” Told he occasionally has a hot-take win: “alll of them;)” Someone says a company he likes will disappear within a year: “will definitely be around. dominating.” Criticism of another prediction gets “stupid. zero chance.”
There is something substantive underneath the swagger. Rabois clearly believes forecasting is a useful test of whether someone actually understands a system. Opinions matter less if you never force them into predictions with outcomes.
But May also shows the weakness of that style: the criteria for “winning” can become elastic. If the precise definition of the Iran war being “over,” or which inflation measure settles a bet, isn't explicit beforehand, prediction markets degenerate into rhetoric. His confidence is informative about his posterior, but not necessarily about whether the proposition was cleanly falsifiable.
That tension—Popperian-looking prediction combined with lawyerly argument over what counts as falsification—runs through the account.
The founder doctrine: ignore everybody and work harder
The clearest serious advice of the month comes from a Shane Parrish quote about resisting instant gratification and having the conviction to pursue something for six months before proving critics wrong.
Rabois distills it:
Put the blinders on and ignore the noise.
He “highly recommend[s]” this to founders.
That becomes especially interesting at the end of the month, when he gets into a long argument about seven-day-a-week companies.
A critic points to failed companies with extreme work cultures and suggests that seven-day workweeks have a bad track record. Rabois immediately reaches for counterexamples:
SpaceX PayPal Apple's iPhone team Uber under Travis Kalanick
His claim is not merely that working hard helps. It is much stronger:
The historical track record of extreme work intensity at genuinely ambitious companies is extraordinary.
When someone argues that SpaceX is different because it was building a category-defining product, whereas companies like Traba are not, Rabois rejects the premise. Supply chain, he says, is “more critical than almost anything we do.”
When someone points out that successful people needn't make overwork their online identity, Rabois partly agrees—but insists that they nevertheless outwork everyone. His examples expand beyond startups to Michael Jordan, Kobe Bryant, politicians, and investors. He credits Donald Trump's political success partly to simply working harder than opponents and attributes Vinod Khosla's longevity at the top of venture capital to the same trait.
His analogy is revealing:
if you want to be Jordan or Kobe, you need to work like they do.
This is the elite-performance model of startups. A startup isn't a normal company with somewhat higher variance. It is professional championship competition. Therefore the relevant comparison isn't the median productive worker; it's Kobe.
The harshest exchange comes when a founder with young children distinguishes personally working seven days from forcing employees into the office seven days. Rabois replies:
let’s see the results.
That crystallizes his philosophy almost brutally. He isn't claiming a priori that every extreme-work founder wins. He's saying the outcome is the arbiter.
When asked whether years of seven-day work were worth it for employees if the result was generational wealth, he answers:
should be.
This is probably the most revealing cluster of the month because it exposes a deep disagreement about what employment at a startup is. Critics are evaluating work practices partly as conditions workers should reasonably expect. Rabois evaluates them through the lens of whether they maximize the probability of an extraordinary outcome.
Those aren't merely different opinions about office attendance. They're different objective functions.
Career advice: optimize for slope, not prestige
Rabois is unimpressed by an elaborate AI-generated ranking of AI companies based on synthesized career advice from himself and several other prominent tech figures:
not very useful results.
Asked what should actually rank highly, he says:
Ramp would be best.
Asked whether Lovable would be better for someone's career than Microsoft or Amazon:
Lovable would be much better for your career.
This is classic Rabois.
The conventional prestige ordering says Microsoft and Amazon are gigantic, selective, globally recognized technology companies. His ordering favors younger, rapidly scaling organizations where a talented person can acquire scope and responsibility quickly.
The implicit model seems to be:
career value ≈ quality of people × rate of company growth × personal responsibility × equity upside
rather than:
career value ≈ current employer prestige + compensation + stability
The irony is that the AI-generated framework actually included several ideas compatible with this—talent density, stage, learning rate, ownership, wave riding—but Rabois rejects the resulting abstraction. That fits another pattern in his thinking: judgment beats frameworks once frameworks become detached from concrete companies and people.
He doesn't answer “apply these eight criteria more carefully.” He names Ramp.
AI: vertical applications survive, generic wrappers don't tell the whole story
Rabois's AI comments are more discriminating than the simple “foundation models eat everything” thesis.
He reacts enthusiastically to Harvey's work with Baseten on post-trained open-weight legal agents. His takeaway:
Legal, accounting and investment banking should thrive as vertical applications of AI.
That's an important claim. He's betting that domain-specific software retains value even as models improve because the valuable product isn't merely model access. It includes workflow, proprietary data, training, evaluation, deployment, integration, economics, and domain knowledge.
The quoted Harvey material reinforces exactly this:
lower inference cost and latency from open weights, visibility into model reasoning, * ability to customize training and architecture.
Rabois then plugs Factory as infrastructure these companies should run on.
When someone predicts Factory will vanish or be acquired because its revenue is supposedly tiny compared with Anthropic, Cursor, or Cognition, Rabois says Factory is “dominating,” its revenue is “much much higher,” and it is accelerating faster than Cursor and probably Cognition.
Whether that claim proves correct isn't available from this corpus. What's useful is what he is implicitly betting on: there remains an important infrastructure/application layer above raw frontier models, even in a world where the frontier-model companies are enormously powerful.
He also likes NavigateAI's pitch of AI copilots for field workers:
Build faster and less expensively!
Again the attraction is straightforward: AI as a way to attack an actual bottleneck in the physical economy rather than another abstract chatbot.
The funniest AI tweet isn't his
Matt Turck posts a satire of the 2026 venture capitalist:
tell portfolio companies to buy more Anthropic/OpenAI, look for startups outside Anthropic/OpenAI, conclude you should invest directly in Anthropic/OpenAI, diligence startups by asking Claude whether it plans to build their product, help portfolio companies upgrade their Anthropic tiers, invite more foundation-model people onto the podcast.
Rabois's entire contribution:
:)
That's enough.
The joke works because it captures a real anxiety in venture capital: if frontier labs capture most of the value, are VCs funding elaborate token-reselling intermediaries while transferring their own capital to model providers?
Another person responds that anyone in San Francisco having this conversation is “literally a year late.” Rabois calls that “quite sharp.”
His reaction suggests he thinks the obvious “AI wrappers are doomed” discourse itself may now be stale. The frontier has already moved to asking which application companies possess enough differentiated workflow, data, distribution, or execution to matter anyway.
Opendoor: distinguish the interface from the business model
Several replies concern Opendoor and agents.
Someone compares a new Opendoor agent initiative with “Key Agent,” which Rabois apparently disliked. He insists:
this is totally different.
The questioner proposes the distinction: rather than pushing consumers into an exclusive agent experience, this looks like a democratized portal layered onto an existing platform.
Rabois agrees:
yes it is a simple portal/API.
When someone interprets pro-agent language as contradicting an apparent thesis that agents disappear, Rabois answers:
time horizon. this is the same as our consumer product now.
That's a useful little window into how he thinks about product transitions. A company can accommodate today's intermediary while building toward a world in which the intermediary's role changes dramatically. Supporting agents now is not necessarily a repudiation of long-run disintermediation.
This is a recurring mistake in technology analysis: treating intermediate product states as declarations about the terminal market structure.
Venture capital as apprenticeship
The warmest material of the month concerns Pierre Lamond.
Others describe Lamond as a legendary investor who trained generations of Sequoia partners. Rabois adds that he trained Roelof Botha as well.
Then:
highlight of my professional life was learning that he complimented me.
That comment is unusually revealing because it comes from someone whose online persona is almost aggressively impervious to external validation.
Apparently Pierre Lamond's validation counts.
When another investor says Lamond is insufficiently studied and mentions his extraordinary late-career investing, Rabois replies:
i have some great stories for you.
This fits Rabois's broader model of Silicon Valley knowledge: great investors and operators transmit judgment through apprenticeship, stories, standards, and direct observation more than through explicit frameworks.
It's the same reason his response to the elaborate Claude-generated career rubric is basically “meh; Ramp.”
The important knowledge is tacit.
His Silicon Valley is intensely personal
Another tiny reply illustrates this.
Asked whether he knows Kevin Warsh personally:
yes since 1988.
Four words, but they convey something about Rabois's information environment. His world consists heavily of long-lived networks of founders, investors, politicians, executives, financiers, and operators.
This matters when interpreting his confidence. Sometimes he's merely opinionated. Sometimes he may possess private contextual knowledge unavailable to the person arguing with him.
Twitter makes those two situations look identical.
Miami as a political case study
Asked to identify well-run Republican cities, Rabois says:
Miami.
When challenged that Miami and Miami-Dade now have Democratic mayors, he credits Miami's quality to sixteen years under a GOP mayor and predicts that the city's institutional culture is now relatively resistant to “stupid Democratic ideas.”
The important idea isn't the partisan insult. It's the institutional hysteresis argument:
A sufficiently long period of competent governance can alter a city's trajectory, norms, constituency, bureaucracy, and economic composition enough that subsequent elections don't immediately reverse it.
He treats Miami not merely as a nice place under a particular politician but as a political-economic system that has accumulated path dependence.
It's asserted rather than demonstrated here, but it's a more interesting claim than “Republicans govern cities better.”
China risk: the Airwallex thesis resurfaces
Rabois reposts his December attack on Airwallex, in which he argued that its operational presence, personnel, investors, and legal exposure in China created a route through which sensitive American corporate financial data could become accessible to the Chinese government.
His May commentary:
This aged well.
The original argument is unusually detailed compared with most of his feed. It alleges risk through several overlapping channels:
China-based personnel, Chinese national-security obligations, Chinese shareholders, operational access to payment infrastructure, * inadequate disclosure to American customers.
Within this corpus, there is no new evidence supplied demonstrating exactly what “aged well.” So the most that can be concluded from the file is that Rabois believes subsequent developments vindicated his warning.
Conceptually, though, it fits his broader tendency to reason from incentives and institutional control rather than branding or formal headquarters. A company being “Singaporean” on paper does not resolve the question if the people, systems, owners, or legal obligations that matter are elsewhere.
The feed is also a masterclass in deliberate terseness
A large fraction of Rabois's replies would be terrible if judged as attempts at persuasion.
That isn't what they're for.
Examples include:
“what?” “correct.” “no.” “who?” “lol” “facts.” “oh god.” “stupid ass.” * “fine, work w a different VC.”
Someone says that, as a founder, they would chafe at a VC describing the relationship as “family.”
Rabois:
fine, work w a different VC.
There is no attempt to reconcile preferences or broaden the market. That's the point. Selection is part of the product.
The implied message is: Founders who dislike the culture should choose another investor, and he should choose founders who actively want it.
This is more coherent than trying to maximize compatibility with everyone.
The jokes work because the persona is consistent
Rabois's humor is mostly dry self-parody.
Someone congratulates him on a semi-rare correct hot take:
alll of them;)
Someone asks whether his always-right confidence remains intact:
yes my bet is no a slam dunk.
Someone mentions Cooley during an analogy mapping elite law firms to venture firms:
who?
The Cooley joke only works because Cooley is obviously prominent enough for the dismissal to be absurd.
Similarly, after someone proposes:
Wachtell = Sequoia Kirkland = a16z Williams & Connolly = Thrive Sullivan & Cromwell = General Catalyst * Latham = Accel
Rabois can't resist correcting the taxonomy: S&C is definitely not General Catalyst, Benchmark is probably Wachtell, and Founders Fund is Quinn Emanuel.
This is peak insider content: turning elite professional-services firms into venture-capital personality types.
The mapping itself is less important than the instinct. He experiences institutions as having very distinct cultures, and he cares enough about those cultures to object when someone gets the analogy wrong.
Sports supply some unfiltered fandom
“OMG. Brunson.”
That's the entire tweet.
Later someone predicts the Spurs winning in five, invoking 1999.
Rabois:
lol. zero chance of that.
These aren't intellectual contributions. They're useful precisely because they break the investor-politics-AI stream and reveal the same personality applied to sports: immediate opinion, maximal confidence, minimal prose.
What ties the whole month together
Rabois's feed can look like disconnected provocations because the topics change constantly. But there is a fairly consistent theory underneath them.
1. Outcomes dominate narratives
Did the company win?
Did the prediction resolve correctly?
Did the founder build something exceptional?
Did the employee receive a generational outcome?
Everything else is downstream.
2. Exceptional outcomes require abnormal behavior
Kobe didn't optimize for balance.
Neither did SpaceX, PayPal, early Uber, the iPhone team, or—as Rabois sees it—the best investors.
Therefore telling a startup to behave like a normal employer may be category error.
3. Judgment is more valuable than elaborate frameworks
Claude can synthesize eight career principles.
Rabois would rather tell you to join Ramp.
The abstraction is useful only insofar as it reproduces expert judgment in concrete cases.
4. Institutions have cultures that compound
Miami after years of Republican governance, Sequoia under Pierre Lamond's influence, elite law firms, great startup teams: Rabois sees organizations as repositories of accumulated standards and behavior, not just collections of people.
5. Time horizon resolves many apparent contradictions
Agents can matter today and be less important tomorrow.
A company can look small now and be “dominating” if its acceleration matters more than its present size.
A career choice that looks less prestigious now can produce much greater future optionality.
6. High conviction is itself a competitive advantage
Founders need enough conviction to spend six months proving everyone else wrong.
Investors need enough conviction to make bets while the consensus disagrees.
Politicians, athletes, founders, and investors win partly because they keep going harder and longer than competitors.
This is probably the deepest Rabois theme.
Why this month is interesting
The interesting thing is how little distinction Rabois draws between startups, investing, athletics, politics, and personal career strategy.
They are all versions of the same game:
- Find a domain where the outcome matters enormously.
- Develop unusually good judgment about what will work.
- Concentrate on a small number of exceptional people or opportunities.
- Ignore social pressure and consensus when your judgment disagrees.
- Work harder than competitors.
- Make decisions early enough that being right actually matters.
- Let results settle the argument.
That framework explains his admiration for Pierre Lamond, Ramp, Vinod Khosla, Kobe, Trump, SpaceX, PayPal, intense founders, and companies he thinks the market misunderstands.
It also explains what his critics dislike. If you believe this model too strongly, survivorship bias becomes extremely seductive. Every extreme practice can be justified by pointing to an exceptional winner that used it. Every losing prediction can potentially be reinterpreted through a different time horizon or metric. And “the results will tell us” doesn't help very much if you haven't specified beforehand exactly which results would prove you wrong.
So the strongest reading of Rabois is not “work seven days a week” or “be contrarian.”
It's:
Treat important decisions as bets on reality, develop judgment good enough to make those bets before consensus forms, and organize your life around maximizing the upside when you're right.
The corresponding warning is equally important:
If you make conviction central to your identity, you need unusually rigorous rules for recognizing when you were wrong. Otherwise the trait that lets you resist noise also lets you resist evidence.
May is compelling because Rabois demonstrates both sides of that trade almost continuously.