Summary

Kaz sees Opendoor as a mission-driven, for-profit refounding whose long-term upside comes from fair home pricing, software-enabled operations, and attached services. Harry framed the broader thesis as a software company with both asset-light and asset-heavy models. Kaz calls the bull case “obscene,” says AI can solve home-pricing problems that were difficult three years ago, and expects Opendoor to launch products that may fail.
Kaz has tied the turnaround to unusually concentrated personal and governance incentives. He is limited to a $1 minimum salary, owns zero RSUs, prefers compensation entirely in options, and would not accept the job without Keith and Eric providing board-level air cover. He left “a few hundred million dollars” at Shopify, uses “hard, valuable, fun” as his operating test, and wants Opendoor to become “the most aggressive in-office public tech company.”
Oracle’s roughly $300 billion future-order revelation was valued as though OpenAI’s promise were already dependable cash. The stock jumped about 36%-38% and briefly touched a $1 trillion valuation, yet the customer reportedly generates roughly $12 billion, remains loss-making, and would need extraordinary financing to supply $60 billion annually. Rory’s conclusion: “Do I think the full last $300 billion will be wired in five or six years? No, I very much doubt it.”
The Oracle trade also exposes the market’s willingness to reward AI revenue without resolving its economics. Jason sees GPU hosting as a fungible, potentially near-zero-net-margin activity, while Rory said model owners will probably capture more value than commodity infrastructure providers. The accounting hinge is unknowable asset life: “If you know over how many years you should depreciate the latest NVIDIA chip, then that would be the key question.”
Microsoft and OpenAI appear headed toward an arm’s-length relationship, not a lasting strategic marriage. The interim MOU signals separation, while Rory guessed that a final arrangement could convert Microsoft’s blocking rights and revenue share into a roughly 20%-35% equity position. OpenAI’s escape from Microsoft’s “bear hug” is judged a stunning founder win, but Microsoft must still prove it built AI that matters once privileged access disappears.
AI applications are expanding markets at extraordinary speed while becoming easier to displace. Higgsfield paired a $50 million raise with $50 million ARR, Gamma reached about $60 million, and Jason cited nominal NRR figures of 140%-180% in leading apps; yet competitors now respond in two weeks rather than six to 12 months. Jason’s deliberately hedged stress case is that Anthropic could lose 30%-40% of Claude Code revenue—and perhaps half its revenue—within 12 months if GTP-5 Codex reaches near-parity.
The panel’s cycle advice is to distinguish marked-up ownership from realizable liquidity and take acquisition offers seriously. Jason expects investors to reject $4 billion fund-returners in pursuit of $8 billion, $12 billion, then $24 billion, only to watch some become worthless; Rory remembers the same “winning the lottery and then losing your ticket” pattern from 1999-2000. Meanwhile, Workday’s $1.1 billion Sana Labs deal shows why an acquirable number two can win, while Adobe’s “AI-influenced ARR” claim casts doubt on whether incumbents have achieved genuine AI-native growth.

Deep dive

1. Opendoor is being refounded around mission and aggression

Kaz said he never expected to leave Shopify, which he joined when skeptics considered it a tiny, heavily shorted company destined to fail. Opendoor changed his mind because making homes easier to buy, sell, and own is both an enormous commercial opportunity and a problem that “matters to the future of our society.”
Harry’s pushback—worth keeping—was whether “we’ll figure out how to make money along the way” reflected boom-era thinking. Kaz’s answer was categorical: “Fuck no.” Opendoor is explicitly for-profit and already has promising bets, but “businesses should not exist to make money. Businesses should make money to deliver on a mission.”
Challenged on the meme-stock dynamic, Kaz said he owns only Shopify and Opendoor and does not trade. His claim is that Opendoor is fairly priced for discounted potential in “the single largest market in the world,” not irrationally priced against present cash flow.
The scale comparison carrying his case: Tesla had not exceeded 10% share in any car market until recently, while Opendoor exceeded 10% in multiple markets several years ago. Kaz called the bull case “obscene,” but conceded that realizing it requires “good stewardship and operational excellence and aggressive execution.”

2. Software leverage is a framing, while Kaz emphasizes fair pricing

Harry framed Opendoor as “a software company that happens to have some assets,” arguing that it could operate with both asset-light and asset-heavy models and that leverage would come from software. Kaz acknowledged that the transaction layer is already software-enabled, said it could use significantly more software and AI, and stated that Opendoor would not remain solely asset-heavy. His central long-term leverage claim was that the company should offer a fair price for a home and add valuable services.
Rory decomposed the system into three capabilities: acquiring buyers and sellers, using a large AI brain to value homes, and executing repairs, processing, and resale efficiently. After only “24 hours and 12 minutes” in the job, Kaz was already impressed by transactional operations, while seeing substantial room for more software and AI enablement.
Harry’s hardest objection was home heterogeneity: comparable sales might determine roughly 91%-93% of value, but gardens, slopes, and other “little shit” determine the margin. Kaz said that was right three years ago but wrong today: “There’s a reason why God invented AI,” while human inspectors are “variance-creating machines.”
The long-range product promise goes well beyond valuation. Kaz wants buyers eventually able to return an unwanted Opendoor home, receive support “for the life of that house,” and treat it as a guaranteed asset; sellers without a next property would have Opendoor find one and manage the liquidity problem.

3. Margin must move from buying cheaply to serving customers repeatedly

Kaz rejected a model dependent on acquiring homes at a discount. Opendoor should offer sellers a fair price, sell a good house fairly, earn trust, and monetize value-added services—similar to Carvana’s fair-price transaction surrounded by additional services. Harry named title and mortgage as examples of potentially profitable services.
Shopify supplied Kaz’s economic analogy: customers could buy its software for $1, no seat expansion was required, and only roughly 25% of revenue came from software. Shopify earned the remainder when merchants succeeded; Opendoor should similarly become “the best deal in buying and selling homes” and earn services revenue through superior underwriting and customer knowledge.
Kaz’s sharpest incentive argument was that making all the profit in one transaction forces shady behavior, as with a used-car seller who has only half an hour to monetize the buyer. A long-running network of homeowners, buyers, and sellers can instead offer some products free and monetize others while preserving trust.
He remained non-dogmatic about agents: experts may help some buyers and sellers, but transactions and relationships involving many intermediaries are usually worse. What he insisted upon was “an excellent fucking product”; believing Opendoor can permanently earn 20% margins simply by buying homes cheaper is “straight up dumb” because markets eventually clear.

4. Options and entrepreneurial board cover align the turnaround

Kaz would take less than his mandated $1 salary and believes corporate executives should “basically only get paid in options.” He criticized structures that can reward executives merely for being “inoffensive enough not to get fired.” He owns zero RSUs, while his package uses option-like instruments and time- and price-based vesting cliffs. Harry referenced a $30 threshold, to which Kaz replied that it represented “a lot of money.”
He also said he will not keep Yahoo Finance on his laptop or watch the daily share price. The intended alignment is long-term user and shareholder value, based on his belief that markets misunderstand Opendoor’s opportunity “by order of magnitude.”
Keith and Eric were non-negotiable: “Wouldn’t have taken the job without them.” Rory described the change as a refounding that needs entrepreneurial air cover while the company absorbs painful decisions; without it, a conventional public board may approve ambition and then retreat when execution becomes frightening.
Kaz’s board model is unusually operational: one director was scheduled to review every owned house line by line, while Kaz had already reviewed every invoice paid over the previous 12 months. Having left “a few hundred million dollars” at Shopify, he organized the company around “hard, valuable, fun” and planned the most aggressive in-office posture among public tech companies.

5. Oracle’s market-cap gain assumes a fragile $300 billion promise

Rory set the scene: despite a slightly light quarter, Oracle disclosed more than $300 billion of remaining performance obligations. The panel inferred that most represented roughly $300 billion of future OpenAI compute, sending Oracle shares up approximately 36%-38%, briefly making Larry Ellison the world’s richest person and Oracle a $1 trillion company.
The bullish arithmetic is coherent if every premise holds: $300 billion over five years equals $60 billion annually, and a five-to-six-times revenue multiple supports roughly $300 billion of incremental market capitalization. Rory’s concern was that the valuation applied “100% certainty” to several uncertain dependencies.
The promised customer was described as generating about $12 billion of revenue, having raised approximately $40 billion lifetime, and still losing substantial money. It may need to raise very substantial additional capital to fund the commitment—Rory mentioned Sam Altman’s $115 billion figure while also saying “a couple hundred billion”—making Oracle effectively a leveraged public-market proxy for OpenAI.
Harry calculated that even if OpenAI doubled by July 2026 and doubled again by 2027, it would reach $48 billion—still $12 billion short of the annual Oracle bill. Rory expects meaningful OpenAI revenue for Oracle but not the entire amount: “Do I think they’re gonna collect $300 billion in orders from OpenAI? Absolutely not.”

6. AI infrastructure revenue may dilute Oracle’s superior economics

Jason’s objection was not merely counterparty risk: nobody appeared to care whether the revenue would be profitable. He characterized Oracle’s GPU hosting as fungible server capacity that consumes cash and could contribute little or nothing to the bottom line for the foreseeable future.
Rory agreed that model owners will probably earn more than commodity cloud suppliers over five to 10 years: he would prefer OpenAI to CoreWeave, and an Oracle business resembling “CoreWeave 2” should deserve inferior economics. That contrasts with Oracle’s existing database franchise and roughly 41% operating margins.
Even positive reported gross margin depends heavily on depreciation assumptions and the usable life of NVIDIA chips. Oracle must commit enormous CapEx upfront for revenue likely to be materially less profitable than databases, leaving the economic result obscured by accounting choices.
The Meta analogy clarified the market signal: investors permit a superb legacy franchise to recycle cash aggressively into AI. Meta was not punished; Oracle was actively rewarded. Rory called the move frothy, while noting public investors were merely joining private AI markets that had already suspended conventional discipline.

7. Stargate’s real product may be momentum and negotiating leverage

Jason reinterpreted the awkward Stargate White House photograph of Sam Altman, Masayoshi Son, and Larry Ellison as an investable signal everyone missed. Oracle’s deal was not separate from Stargate but evidence that “this is Stargate”; in retrospect, he said, “We should have bought Oracle stock that day.”
Rory’s framing: every participant wins before the compute is delivered. Oracle receives an enormous valuation uplift, while OpenAI demonstrates to Microsoft that another provider will fund vast CapEx, strengthening Sam’s negotiating leverage.
That creates incentives to announce a credible possibility without proving it will happen. For an ambition as large as OpenAI, momentum is both friend and necessity because it compounds inevitability; investors were supposed to ask whether $60 billion of annual revenue was believable and “clearly, they forgot to do that this week.”
One panelist said venture had shifted from investing toward trading—pay a high price and hope someone pays a more irrational one. Jason thought the game might have “a couple good years” left; Rory invoked Chuck Prince’s 2007 warning that while the music plays, participants keep dancing.

8. This is the cycle to accept fund-returning liquidity

Jason expects a defining mistake over the next 24-36 months: boards will reject $4 billion offers for AI businesses with 3% gross margins because they want $8 billion, then $12 billion, then $24 billion. Some will awaken to assets worth nothing.
Rory had seen the movie in 1999-2000, when boards rejected $2 billion for fiber-optic or communications businesses while demanding $8 billion. Meeting those teams after the companies reached zero felt like “winning the lottery and then losing your ticket.”
One panelist’s distinction was valuation versus liquidity. A higher private mark does not mean an investor can exit; available secondaries may impose discounts, limit sellers to 10%-20% strips, and prevent full realization. Public liquidity distributes declining prices among successive owners, whereas private investors may hold the same position all the way down.
Jason now tells every founder with a major offer to take it, partly to force truly confident founders to answer, “No way. This is gonna be bigger.” Rory disliked the blanket wording but agreed that record valuations change the Bayesian prior; founders selling near prices VCs just offered, as discussed around Scale AI and Windsurf, may know more than incoming investors.

9. Microsoft and OpenAI are consciously uncoupling

Jason noted Microsoft was already shifting parts of Office toward Anthropic, making it the default for several products and telling teams months earlier to use Claude Code. OpenAI may gain a lower revenue share and freedom to partner elsewhere, though the “price in blood” paid back to Microsoft remained unclear.
Rory stressed that the announcement was an interim MOU, not a final agreement. Microsoft’s unusual investment carried profit participation, revenue share, and blocking rights that became a “poisoned chalice” for OpenAI’s emergence as a normal standalone company.
Rory’s guessed settlement would convert Microsoft’s roughly $13 billion investment into a 20%-35% equity position worth perhaps $100-$150 billion if OpenAI were valued at $500 billion. He also described a possible $100-$200 billion value range for Microsoft’s stake. That is an exceptional venture return but insufficient to transform a roughly $3 trillion company; the larger benefit was Microsoft’s two-year AI credibility and market-cap lift.
In three years, Rory expects a normal relationship: Microsoft remains a large shareholder, sells non-exclusive Azure capacity, and buys non-exclusive model access. Anthropic and OpenAI both used their Amazon and Microsoft alliances for money, compute, credibility, and critical mass, then achieved independence—Sam effectively escaped “one of the greatest bear hugs of all time.”

10. AI applications expand TAM faster than old spreadsheets can capture

Higgsfield announced both a $50 million raise and $50 million ARR in a timeframe presented as faster than Lovable or Replit; Gamma reached about $60 million from nearly zero that year. Jason’s point was that short video and slides can become enormous AI markets even when they look niche from traditional software assumptions.
Rory described a 10X, perhaps 100X, expansion in access: ordinary internet users can now code or create videos that previously required trained specialists. Jason sharpened it—work that was impossible six or seven months earlier can now be done “for pennies.”
That invalidates “TAM spreadsheets from 2021.” Gamma, Higgsfield, and Opus Clip do not merely steal existing professional spend; they enable new volumes of content. The opportunity is therefore larger, even if the applications sit atop models they do not control.
Harry’s concern resembled COVID-era forecasting: which growth is enduring and which is experimental, whimsical behavior that disappears after one cycle? Jason pointed to nominal NRR of roughly 140%-180%, while conceding durability risk; he also distinguished cash-flow-positive Higgsfield from Replit and Lovable’s negative margins.

11. Two-week competition makes even foundation-model revenue fragile

Jason’s tentative durability test had two parts: enough white space to keep expanding with customers, and a founder “maniacally focused” on building adjacent products. Lovable and Replit may surround website creation with every ancillary service needed to operate the resulting site.
The competitive window has nevertheless collapsed. Jason said startups once enjoyed six to 12 months before rivals decided, built, and responded; now the lag can be two weeks. Rory’s unsentimental answer was that everyone dislikes competition while refusing to leave an opportunity compelling enough to attract it.
Jason’s deliberately conditional stress case reached the model layer: Anthropic could lose half its revenue over 12 months if GTP-5 Codex became as good as Claude Code. Even at 95% parity, switching inside Cursor, Lovable, Replit, and other applications could conceivably remove 30%-40% of Claude Code revenue.
Harry hears two simultaneous truths from decacorn CEOs: they are so price-insensitive toward Claude Code that they would spend 10X, yet would switch tomorrow for a comparable product. Rory connected that instability to “insatiable” CapEx—more pre-training and reinforcement learning may be the only defense, creating individually rational behavior that aggregates into something frightening.

12. Distribution and acquirability determine incumbent outcomes

Wix offered the clearest “revenge of the incumbent” specimen. It bought Base44, described as an $80 million, solo-founder Lovable/Replit clone, then added safety, identity, and its distribution funnel; the business was projected near $50 million ARR within months. Jason noted that even a 10% market share would be significant for such a rapidly distributed product.
Adobe and Salesforce have launched AI products but not generated equivalent explosive growth. Scale matters: $100 million can move Wix, while it barely registers against Adobe’s roughly $23 billion or Salesforce’s $40 billion-plus revenue base.
Harry challenged the idea that number one is always unacquirable by pointing to Scale AI’s $14.9 billion sale. Jason replied that hyperscalers can pay at that level, but a buyer such as Workday cannot pay enough to make Glean’s investors whole.
Workday’s announced $1.1 billion purchase of Sana Labs, discussed at roughly $50 million ARR, illustrated Jason’s “number two” thesis. When Glean or another leader is unavailable or unaffordable, the runner-up receives strategic offers—provided it has not raised so much that acquisition cannot satisfy investors.
Bending Spoons offered the cash-flow alternative, buying flat, roughly $420 million-revenue Vimeo for $1.38 billion, around 2.5-3 times sales. Harry questioned whether its Evernote playbook—centralize, cut costs, raise prices—works as easily at this entry price; growth or much greater profitability is now required.

13. IPO issuance reopened, but the underlying assets sharply diverged

Jason called it the busiest IPO week since 2021 and a meaningful micro-milestone: a market that decelerated had recovered its old pace for one week. Harry welcomed the liquidity but noted LPs may need roughly 12 months to receive cash that can recycle into venture funds.
Rory found Figure the most interesting. Mike Cagney’s second act raised close to $800 million while using blockchain to settle, process, and securitize home-equity and other non-conforming loans—“finally a use for the fricking blockchain” beyond trading—though the company will still rise or fall on underwriting quality and credit losses.
Gemini was introduced at a $4.4 billion valuation and a 32% first-day bump, but Rory noted it surged and then retreated intraday while revenue was declining. It remained, in his view, another crypto exchange whose differentiation from Coinbase and Binance was unclear.
Via was introduced around $3.5 billion, while Jason cited a roughly $4.2 billion valuation after $493 million invested across 13 years—approximately 10X aggregate capital. He called that A-plus but perhaps no longer S-tier; Rory would “take it in a heartbeat.” The shares initially opened below the offer price before recovering.

14. OPEN trades on momentum while Adobe confronts seat cannibalization

From OPEN at $9.30, Jason predicted $24 by December 31, extrapolating Kaz’s quality and online memer enthusiasm. Rory expected something between $9 and $24 because Kaz can create momentum and make the story feel enormous, but remained skeptical that a business full of “special snowflake” houses can create massive enterprise value over three to five years.
Rory called Adobe a profitable, slow-growing company that had “returned to Earth,” around five-to-six-times sales and roughly 15-times forward earnings after previously reaching about 18-times revenue. His low-confidence 12-month ceiling was approximately 10% upside if revenue also grew 10%.
Jason instead predicted at least 10% downside. Scott Belsky’s departure was one warning; Adobe’s claim of $5 billion in “AI-influenced ARR” was another: “You don’t have to say this if you have AI ARR.”
The structural trap is the seat model. Adobe wants Firefly to defend an expensive suite without cannibalizing licenses, while Higgsfield or Gamma may require only one seat—or no conventional seat at all. With Canva, Figma, and a new generation democratizing creation, Rory concluded that believing AI does not fundamentally threaten Adobe would be “delusional.”

No Priors Ep. 97 | With Decagon CEO and Co-Founder Jesse Zhang E207 | Agent Founders’ Cost Breakthroughs and Commercialization