Keith Rabois
The month in one sentence
Rabois spent August 2026 doing three things with unusual consistency: defending Opendoor with absolute conviction, advertising a hard-edged founder/VC operating philosophy, and treating Twitter as an arena where being concise, combative, and publicly certain is itself part of the argument.
The interesting part is less any individual opinion than the coherence of the worldview underneath them: business outcomes matter more than accounting conventions, elite judgment matters more than consensus, speed matters more than etiquette, and people who repeatedly fail his judgment test are not merely mistaken but “clowns,” “losers,” or “fools.”
Opendoor: the dominant obsession
By far the biggest thread was Opendoor. Rabois spent days fighting shareholders, critics, journalists, and random accounts over whether the company is fundamentally improving.
His case is remarkably simple:
Previous management was disastrously incompetent. The company deteriorated under that regime to a valuation below roughly $250 million. The new regime cut dramatically while still producing roughly 5–5.5× growth, according to Rabois. Marketing spend supposedly fell from something like $80 million to $5 million while growth accelerated. The company remains on its promised path toward positive ANI/profitability. Therefore, judging management by the current stock price is missing the operational turnaround.
When someone complained that bulls had suffered nothing but losses, Rabois answered that the company had been worth under $250 million before the prior CEO was removed: “QED.” When another invoked the old $20 share price, he said the premise was false and predicted the company would eventually beat that valuation.
He repeatedly challenged skeptics to actually short the stock. Near month-end, when someone called him a useful contrarian indicator, he replied:
“please short. hopefully with all of your assets.”
And after the person pushed back, Rabois ended with:
“the company is in great shape and you will learn the hard way.”
Why this is interesting
Rabois isn't merely defending a portfolio company. He's exposing his preferred way of evaluating one.
He cares much more about operating trajectory—growth, cost structure, inventory quality, future profitability—than the market's current verdict. His implicit model is:
share price is noisy; operational transformation is causal.
That is a legitimate and very venture-capital-style way to think, but Twitter turns it into an unusually pure test of conviction. He doesn't hedge with “assuming execution continues.” He makes falsifiable predictions: the valuation will recover, profitability will arrive on schedule, skeptics will lose.
That makes this cluster worth revisiting later because reality can eventually score it.
The Deirdre Bosa fight: GAAP versus how investors should actually think
A second Opendoor argument became a broader argument about financial analysis.
Journalist Deirdre Bosa reiterated that her disagreement with Rabois was mundane: if stock compensation is an expense, you cannot exclude it and then claim GAAP profitability. “The GAAP math is the GAAP math.”
Rabois's response was basically: that is exactly the wrong way to think about a business.
He said entrepreneurs are explicitly counseled to avoid obsessing over GAAP and “silly EBITDA fictions” and instead report cash flows. Elsewhere he defended the principle that a company can rationally remain unprofitable if reinvesting capital creates more valuable future assets.
What escalated this from accounting disagreement into characteristic Rabois theater was his treatment of Bosa personally. Across several replies she became a:
“clueless clown” “fool” “stubborn” fool “stupid entitled journalist” * example of why “legacy media is dying”
When someone argued that publicly refusing to let a polite disagreement die reflected badly on his judgment, Rabois answered:
“i explained how business works. she didn’t want to learn.”
Why this is interesting
There are actually two arguments tangled together.
The narrow accounting point is mostly definitional: GAAP profit means GAAP profit.
Rabois's more substantive claim is different: GAAP profitability may be a poor measure of whether management is allocating capital intelligently or creating shareholder value. Cash flow, unit economics, growth efficiency, and reinvestment opportunity can matter much more.
Those claims are compatible. Much of the fight comes from Rabois refusing to accept the narrower framing because he thinks it teaches people the wrong mental model.
It also reveals an important feature of his epistemology: once he concludes that someone is reasoning from the wrong abstraction, he stops treating the disagreement as reasonable pluralism and starts treating it as evidence of incompetence.
His management philosophy: manufacture discomfort
Several quieter posts give a much clearer picture of Rabois as an operator.
He endorsed Frank Slootman's idea that:
“Leaders Must Manufacture Discomfort.”
He also strongly agreed with Scott Belsky's argument that consulting more people tends to produce worse or more average decisions.
These fit together neatly.
Rabois appears to prefer organizations in which:
decision-makers exercise concentrated judgment; leaders deliberately prevent complacency; consensus is suspicious; speed beats inclusiveness; * extraordinary people should be trusted more than process.
That same philosophy appears in his response to a claim that early-stage founders shouldn't need executive assistants.
Rabois initially resisted hiring an EA for roughly his first decade in tech. Jack Dorsey eventually insisted he hire one at Square, and Rabois now says Dorsey was “100% correct.”
His explanation is unusually good:
not having one is imposing costs on the rest of the team.
In other words, optimizing your own apparent self-sufficiency can be globally inefficient. If everyone else must bend around your scheduling, travel and communication constraints, your refusal to delegate creates distributed organizational overhead.
Rabois says he's unusually strong at managing his own time—a founder apparently calls him the “world's best EA”—yet still concluded that refusing assistance was wrong.
He wants to discuss calendar audits and time-management techniques, which he calls some of his favorite founder discussions.
Why this is interesting
This is Rabois at his most useful because the claim is counterintuitive without being merely provocative:
personal efficiency and organizational efficiency are different optimization problems.
A founder might be perfectly capable of scheduling meetings but still destroy more collective productivity by doing it personally.
That same lens explains a lot of his other thinking: optimize the system, not the socially obvious metric.
Venture capital: judgment compounds, networks decay
One of the month's best exchanges came from Ben Casnocha, who argued that venture firms can simultaneously experience:
declining networks as partners age, retire or lose relevance; strengthening brands as old successful investments become increasingly famous.
That can push established firms toward later-stage investing: the brand still wins competitive deals, but the network is less effective at discovering unknown founders at day zero.
Rabois simply replied:
“astute.”
That brevity is noteworthy because this is precisely the kind of VC observation his own career makes salient.
Elsewhere he pushed back on an assertion that today's leading firms were mostly new, saying at least 13 of 18 had existed for twenty years.
He also rejected an analogy between Legora/Harvey and Ramp/Brex. His argument was that Brex had obvious defects and he recognized Ramp as a winner “day 1.” If Harvey has defects, he suggested, Legora is copying them—and he is highly skeptical anyone running an applied-AI startup is a better CEO than Harvey's Winston.
That last point is pure Rabois: markets may look like product races, but he constantly reduces them to founder quality.
Raising too much money can reduce optionality
He strongly agreed with an argument that founders often misunderstand fundraising.
The conventional intuition is:
more cash = more options.
The counterargument he endorsed is:
more cash → more spending → not necessarily more progress → higher expectations and valuation → fewer viable financing paths later.
After eighteen months, a heavily funded startup may have achieved roughly the same milestone as a leaner one, but it can no longer plausibly raise a modest bridge or reset into a smaller Series A.
This is a classic example of the kind of inversion Rabois likes: resources that appear to expand freedom can actually create commitments that reduce future degrees of freedom.
AI: the harness matters, routing becomes infrastructure
Rabois's AI posts were mostly commercial rather than philosophical.
He agreed with the idea that model selection should increasingly happen at the agent/harness layer, rather than simply sending everything through one gateway or one model.
The underlying thesis:
every task has a different optimal model mixture; different harness logic changes performance; the Pareto frontier between accuracy, cost and latency continually moves; therefore static model selection is inefficient.
He recommended Factory in that context.
Then he enthusiastically promoted Ramp's Router product, which claims to send each request to the model best suited for the task while controlling token spending. The quoted claim was roughly 40% lower cost for equivalent outputs.
A week later he highlighted Router traffic supposedly compounding at approximately 34% per weekday:
“Boom.”
Why this is interesting
The implicit bet is that models themselves become increasingly substitutable components, while valuable software migrates one level up the stack.
If models constantly leapfrog one another on price, reasoning quality and latency, the durable product may be the system that dynamically decides which intelligence to buy for each request.
That's a materially different future from “pick OpenAI/Anthropic/etc. and build around it.”
He was similarly unworried that an AI accounting startup might get crushed by model companies moving upward into applications. That suggests he sees significant durability in specialized application layers rather than believing foundation-model vendors inevitably absorb everything.
AI as a weapon against professional services
On a post noting the overwhelming Democratic political orientation of elite BigLaw, Rabois joked:
“Another reason to replace lawyers w AI:)”
It is partly partisan trolling, but it sits inside the broader thesis above: Rabois appears enthusiastic about AI attacking expensive professional-service layers.
He also recommended replacing traditional commercial insurance brokers with a newer technology-driven alternative, promoting a company whose founder had previously attacked real-estate brokers with Opendoor.
The recurring pattern is obvious:
incumbent intermediary + expensive human coordination = target.
Ramp: unabashed portfolio-company evangelism
Rabois spends a lot of time talking his book and does not pretend otherwise.
When someone remarked that Stripe owns a large percentage of Ramp, Rabois corrected them: he said he personally led the seed, Series A at Founders Fund, Series C with Nap, and Series F financings, while Stripe's ownership was “immaterial.”
A long back-and-forth followed over what actually appeared on the cap table, including the revelation that someone Rabois expected to see among major holders wasn't even in the top twenty.
He also jokingly tried recruiting someone after they marveled at accelerating growth at ten-figure scale:
“you want to join Ramp? :)”
The month's funniest Ramp promotion came with Router:
“As always, Ramp saves you time & money. Now we save you 40% on your fastest growing and increasingly crucial token spend.”
Subtlety is not part of the strategy.
Enterprise sales and enterprise software
Rabois linked a podcast as:
“Everything you need to know about enterprise sales.”
Another exchange shows the worldview underneath that interest. When someone promoted an open-source coding harness as a competitor to one Rabois favors, he dismissed it with:
“never even considered by enterprises.”
That is an important distinction in his investing framework. Something being technologically impressive, popular with developers, or open source does not automatically mean it possesses the properties that large enterprises actually buy.
This appears repeatedly in his thinking: distribution, buyer behavior, trust and organizational adoption matter as much as product capability.
Geography: the Bay Area is still the Bay Area
A statistic noted that around one-third of the year's VC-backed startups raising at $10B+ valuations were in the Bay Area, while suggesting there were plenty of opportunities elsewhere.
Rabois's response:
“Narrative violation.”
The joke targets the recurring story that startup geography has decisively decentralized away from Silicon Valley.
He also joked that a discussion applied to “Mountain View too :)” rather than only Palo Alto.
Yet Rabois himself describes an almost comically distributed life:
“travel from Miami to NYC to DC to SF every month living everywhere.”
He continues to insist Miami is “the best,” while explaining that he has children in DC.
So his geographic position isn't really “everyone should live in Miami.” It looks more like high-mobility network arbitrage: live where you prefer while maintaining physical presence across the country's important nodes.
Health as another optimization problem
The month opened with Rabois promoting his own “secrets of optimizing health & performance (and intellectual curiosity).”
When challenged, he pointed to his own:
resting heart rate; two-minute heart-rate recovery; * VO₂ max.
Another person joked that creatine trades at higher multiples than product-led-growth SaaS. Rabois replied:
“creatine is better for you.”
And when someone described 7 a.m. as intense:
“7 am is slacking for me:)”
This is partly self-branding, partly genuine operating philosophy. Rabois treats physical performance much as he treats companies: measurable inputs, measurable outputs, continuous optimization.
Politics and foreign policy: absolute confidence, very little calibration
Politically, Rabois remained strongly combative.
When a gay immigrant said Democrats resent his economic success while Republicans threaten his rights, Rabois answered:
“MAGA loves gays.”
No qualification followed.
He dismissed another person's political analysis on the grounds that “Jason has never been correct about anything political in 20 years.”
He also continued defending an earlier prediction about the Iran conflict. Critics called his “4 day Iran war” call a major miss. Rabois repeatedly answered:
“it was correct.”
His defense was that the war was effectively over and that American casualties after action began were low compared with both earlier casualties and historical military operations such as Grenada.
By August 27, when asked simply:
“Is the war over?”
Rabois answered:
“yes”
Why this is interesting
The style is almost anti-Bayesian.
Rabois rarely communicates probabilities, uncertainty ranges or conditional forecasts. He converts predictions into binary claims and then litigates whether the underlying reality satisfies his definition.
That can be powerful when conviction is based on superior information or judgment. It can also make it difficult to distinguish genuine calibration from ex-post reframing.
His feed therefore contains lots of claims that are unusually easy to audit later—which is useful.
Israel/Gaza
When someone challenged him over a claim that children in Gaza were threatened over kite-flying and asked whether this fit Rabois's conception of morality, he did not engage the factual scenario. Instead he answered:
“yes because you are still lying 100% of the time.”
The meaningful thing here is the argumentative move: he rejects the credibility of the source rather than debating the individual allegation.
That recurs throughout the month. Once Rabois classifies someone as dishonest, stupid, trolling, or incompetent, he largely stops evaluating their propositions individually.
Infrastructure and politics
Rabois endorsed a political argument attacking opposition to data-center construction.
The quoted thesis was that politicians trying to stop data centers are simultaneously opposing both future employment and the physical infrastructure underlying services such as Netflix, X, YouTube, Spotify, Google and Amazon.
Rabois:
“Indeed.”
This fits his broader pro-build worldview: economic progress requires physical capacity, not merely software abstraction.
He also endorsed State Affairs in response to pessimism about America's ability to fix K–12 education, arguing that large-enterprise CEOs are irresponsible if they aren't adopting it.
Media criticism
Rabois's contempt for legacy journalism wasn't confined to the Bosa fight.
When someone lamented the replacement of news by entertainment, he referenced Neil Postman's Amusing Ourselves to Death.
His Bosa dispute then became evidence, in his mind, for why legacy media is dying: journalists don't understand how businesses should actually be analyzed, yet insist on teaching simplistic frameworks.
Whether or not the indictment is fair, his model is clear:
the Internet disintermediates institutions whose authority exceeds the competence of their individual members.
That is basically the same thesis he applies to lawyers, brokers and other intermediaries.
A few smaller but revealing exchanges
When someone argued that Larry Ellison starting another enterprise-software company would have perhaps a 90% probability of success while Zuckerberg launching another consumer-social startup might have only 9%, Rabois simply said:
“yes.”
It encapsulates his extreme belief in domain-specific founder edge. Greatness is not infinitely portable.
When asked for a product first adopted by VCs before going mainstream:
“linkedin.”
When someone said having more advisers makes decisions worse:
“This!”
When shown the historical fact that Philips once owned all of ASML and 28% of TSMC while being worth only a tiny fraction of those companies today:
“Wow.”
When someone accused him of having passed on TSMC:
“no actually.”
The feed oscillates constantly between sophisticated pattern recognition and playground-level trash talk.
The comedy
Rabois's humor is almost always one of four kinds.
Extreme understatement
A sophisticated observation gets:
“astute.”
An extraordinary corporate-history fact gets:
“Wow.”
An explosive growth chart gets:
“Boom.”
Status humor
“7 am is slacking for me:)”
Or responding to someone calling him a broken clock by observing that this broken clock somehow achieved “a higher success rate than virtually anyone.”
Portfolio-company propaganda
Someone marvels at huge growth:
“you want to join Ramp? :)”
Insult escalation
A large proportion of the feed is effectively improvised insult comedy:
“stupid ass” “moron” “loser” “clown” “troll” “fool” “third grade level” “stupid entitled journalist”
There is almost no effort to preserve dignity once he decides an interlocutor deserves contempt.
The deeper pattern: Rabois hates proxies
The most interesting commonality across the month is that Rabois repeatedly attacks proxy metrics and institutional proxies.
He dislikes:
GAAP profit as a proxy for business quality; stock price as a proxy for operational progress; amount raised as a proxy for startup strength; consensus advice as a proxy for good judgment; famous VC brand as a proxy for current sourcing network; open-source popularity as a proxy for enterprise adoption; journalistic authority as a proxy for business competence; raw access to AI models as a proxy for having the best AI system.
Instead he keeps trying to descend one level closer to mechanism:
cash generation; growth efficiency; founder quality; cost structure; actual enterprise buying behavior; actual time allocation; actual model-task performance; actual network strength.
That is the intellectually strongest through-line in the feed.
But he has his own proxy: track record
The obvious tension is that while Rabois distrusts most proxies, he puts enormous weight on demonstrated elite performance.
He invokes:
his investing success; identifying Ramp immediately; operating experience; his time-management ability; superior founder judgment; third-party studies supposedly documenting his unusually high success rate.
So the worldview isn't “never trust proxies.”
It's closer to:
trust causal understanding first; when that is unavailable, trust people with demonstrated judgment rather than institutions or consensus.
That is a recognizably venture-capital epistemology.
Its failure mode is equally obvious: someone highly successful in several domains can start interpreting past success as evidence of correctness in unrelated domains.
The Ellison-versus-Zuckerberg exchange shows that Rabois himself understands domain specificity when evaluating founders. His political tweets exhibit much less of that caution.
What changed or mattered this month
The actual informational payload of the month is roughly:
Rabois remains extremely bullish on Opendoor and says its profitability timeline has not changed. He claims dramatic growth and marketing-efficiency improvements under the new regime. He says legal restrictions prevent him from buying Opendoor stock and that the restriction is not simply a normal blackout period. He remains exceptionally bullish on Ramp and is actively promoting its move into AI-spend routing. He thinks dynamic model routing at the harness/application layer will be important infrastructure. He is optimistic about vertical AI applications surviving rather than simply being swallowed by model vendors. He believes founder/CEO quality remains one of the dominant explanatory variables in startup competition. He endorses lean fundraising, concentrated decision-making, organizational “discomfort,” aggressive time management and the selective use of EAs. He still considers Miami the best home base while effectively circulating continuously among Miami, New York, DC and San Francisco. He continues defending his prior geopolitical prediction rather than conceding a miss.
The most useful way to read Rabois
Don't read the feed primarily as a source of settled facts. Read it as a live display of a particular decision-making architecture.
Rabois repeatedly asks versions of:
- What is the actual underlying variable that matters?
- Which conventional metric is distracting us from it?
- Who has demonstrated unusually good judgment here?
- What would an aggressive operator do rather than merely discuss?
- Where is consensus creating complacency?
- What claim am I willing to make strongly enough that reality can eventually prove me wrong?
The great strength of this style is that it produces decisive action and genuinely non-consensus observations.
The great weakness is visible in exactly the same tweets: confidence can outrun argument, contempt can substitute for rebuttal, and an excellent track record can become an all-purpose credential.
That tension—between unusually sharp operating insight and almost comically uncalibrated certainty—is what makes this month's Rabois feed interesting.