The SaaS Massacre: Public Market Collapse |Microsoft Lost $360B & NVIDIA’s $100B Dispute with OpenAI

Summary

• SpaceX completed its acquisition of xAI at a $1.25 trillion combined valuation — Elon “has now bought Twitter three times in four years.” The panel’s frame: SpaceX holders eat 20% dilution for an asset doing “some 4 million in revenue” against SpaceX’s ~$19B growing 30% profitably, but the deal ships with an instant secondary, as understood, and a 25% paper markup over the $800B mark of eight weeks ago — “you don’t like it? Sell.” The rescued party is xAI, “the orphaned little Chucky in the storm,” lashed to the SpaceX mast because “Elon operates like the Marines. No investor left behind.”
• The bigger signal is “the rehabilitation of the IPO — the end of stay private forever.” Harry’s pattern-read: xAI needs capital, Anthropic is seriously planning to IPO this year, and OpenAI is slowing hiring — the only reason to slow hiring is to show declining losses into a listing. Rory’s Jesse Livermore 1904 parallel: the smart money has realized “there’s not infinite money available — I better get mine.” Caveat: the window is only for $4B revenue growing 50%+ — below that line, “everyone except Palantir has been massacred.”
• “Inference is the new sales and marketing.” OpenAI’s core claim — compute and revenue have a 1:1 correlation — means “it makes sense to consume every single penny of capital on all of planet Earth… a perpetual motion machine for the moment.” Lemkin’s venture corollary: inference-as-GTM is “the only play that works in venture today.”
• The SaaS massacre is a durability crisis, not a churn crisis: top-25 public software growth has decelerated every single quarter since Q1 2022 (Atlassian −67% in 12 months, Gartner −71%, Klaviyo −38%), yet churn hasn’t spiked for systems of record — new customer growth simply slowed. The bottom call, verbatim: “You don’t see a bottom until these things are at free cash flow multiples, net of dilution… and when that happens, that’s your bottom. What a shitty time.”
• Venture has gone binary: private companies are “growing at insane rates or unfundable,” public ones accelerating and not decelerating — “it’s are you growing like a beast,” and you now get “22 months to prove yourself in venture.” Lemkin “smells decay” in everything sub-hypergrowth; Rory’s dot-com pushback: some 10x growers will flame out on bad economics while 2–3x compounders quietly win — “don’t assume the former will entirely track the latter.”
• Microsoft’s $360B one-day loss came from a 1% Azure miss (37% vs 38%) plus doubt that the 40–50% of RPO tied to OpenAI is money-good. The diagnosis: corp dev executed brilliantly (a ~30% OpenAI stake), the product team didn’t — no model, no compelling AI apps — and over the medium term Microsoft, which once took “70–80% of all the profits made in software,” must own the model layer or watch OpenAI and Anthropic each do $50–100B in software revenue.
• The NVIDIA–OpenAI $100B spat is a growth-derivative problem, not a solvency one: the joint press release said NVIDIA “intends to invest up to $100 billion,” Jensen now reframes it as merely “offered,” and Sam hit back that “your chips are too slow.” The panel rejects “too big to fail”: the risk is promising 10x and delivering 5x, rippling through Oracle and the circular economy. One guest still “wouldn’t be shocked” by government-backstopped 0% data-center financing within 24 months; another calls it “a mistake and toxic” — “you’ve built 100 data centers, we only need 50. There’s no point in guaranteeing another 50.”
• Waymo’s $16B raise at $110B (~$350M run rate, 3x oversubscribed) is “cheap” by Tesla decomposition: strip a 2x multiple on the flat car business from Tesla’s $1.2T and the market prices Tesla’s zero-commercial-revenue self-driving at ~$500B — Waymo is a working program at 20% the price. But ~80% of any Elon asset is “Elon premium,” and he is “the most valuable human in history… one horrible car crash away from $2 trillion of value destruction.” Meanwhile Moltbook’s 1.5M connected agents were “fake — we’ve been punked,” yet agent-to-agent communication just happened at scale, “one skill MD away from pretty nefarious stuff.”

Deep dive

1. SpaceX buys xAI at $1.25T — Elon purchases Twitter for the third time

• The news dropped hours before recording: SpaceX completed its acquisition of xAI, valuing the combined private company at $1.25 trillion. Harry’s framing: he’s done this before — SolarCity mashed into Tesla when it “wasn’t going to make it,” Tesla cash lent into xAI — “load balancing his portfolio across the greater outcome and the greater good. We just don’t all have this luxury.”
• Jason Lemkin’s favorite line from Twitter: “Elon has now bought Twitter three times in four years. He bought it standalone, then he bought it at X, and now he’s bought it at SpaceX… he really likes that product, baby.”
• Rory’s two-bucket framework: first the industrial logic — do these belong together? — which forces “a developed opinion on the economics of data centers in space,” since Elon’s “wild” deal note leans heavily on ascending the Kardashev scale and harnessing “a non-trivial amount of the sun’s power.” Second, the margin question, phrased so it answers itself: “Would you prefer to be a SpaceX investor taking 20% dilution here, or a Twitter/xAI investor rolling into the largest market cap private company on the planet maybe 6 months before it goes public?”
• On firepower: “The only person of whom you can truly say he has unlimited cash is in fact Elon” — but at the margin the combined entity raises money more easily than xAI alone, even after xAI’s raise at “200-something billion pre.”

2. The merger math: 20% dilution, softened by an instant 25% markup

• Harry’s devil’s advocate: why would a SpaceX employee want this dilution? Lemkin’s answer — the structure insulates the sting: an instant secondary for everybody, as I understand it, and an instant markup, SpaceX having been worth $800B “a couple shows ago.” “On paper our share price has gone up 25% and we can sell. Who gets to complain? You don’t like it — sell for a material markup from 8 weeks ago.”
• Rory, self-described Debbie Downer: you owned 100% of SpaceX to hold for a decade; now you own 80% of SpaceX and 20% of something else. On revenue multiples the 80/20 split looks lopsided — SpaceX at ~$19B growing 30%, profitable “allegedly” (“we haven’t seen the data”), the other side doing “some 4 million in revenue” as he put it.
• His own resolution: that’s “playing small ball.” If the industrial logic is real, “being wrong 5% on price doesn’t matter at some level” — which is why the space-data-center thesis has to carry the deal.

3. “The rehabilitation of the IPO” — the end of stay private forever

• Harry’s pattern-read across three simultaneous headlines — the SpaceX/xAI deal, Anthropic “seriously planning to IPO this year,” and OpenAI reportedly slowing hiring: “there’s subtle pressure around access to capital. They’ve all got to have an IPO strategy sooner than they’d hoped.” His hiring-slowdown theory: OpenAI wants to “IPO with massive losses but show declining losses more quickly than they planned. Why else would you slow hiring?”
• Rory’s headline call: “What you just saw is the rehabilitation of the IPO. I’m going to call it the end of stay private forever… We’ve now found all the private capital on the planet. It’s still not enough.” Soon “there’s going to be banking teams with guns against their head told: start planning, we want to get all these things public.” His 120-year reference: Jesse Livermore in Reminiscences of a Stock Operator, watching companies pull forward capital raises in 1904 — “the smart money’s realized there’s not infinite money available, I better get mine.”
• The deal makes a SpaceX IPO harder (more complex story, losses) but — “Elon operates like the Marines. No investor left behind” — it saves xAI, “the orphaned little Chucky in the storm,” by lashing it to the SpaceX mast.
• Lemkin’s crucial caveat: this rehabilitation is for Anthropic/OpenAI/SpaceX-scale names only, “not for cute CRM for doctors in certain geographies.” The new bar for a good IPO: “four billion growing 50% or more — that’s EquipmentShare or above.” Below that line, “everyone except Palantir has been massacred.”

4. Compute equals revenue — “inference is the new sales and marketing”

• The engine under all of it, per the panel: OpenAI keeps saying compute and revenue have a 1:1 correlation. “As long as that holds, it makes sense to consume every single penny of capital on all of planet Earth… at least for topline, it’s the greatest money-making machine ever generated in the history of mankind. It’s a perpetual motion machine for the moment.”
• Rory’s analogy: SaaS had a decade where sales-and-marketing spend worked exactly this way — dollars in, more dollars out. Microsoft said the same on its call: compute-limited, but “they could turn compute into money at the drop of a hat.” And if, as Elon says, you can’t build in Tennessee, “then by God you’ll build in low Earth orbit.”
• Lemkin’s founder-facing distillation: “Inference is the new sales and marketing… You either grind it out with thousands of reps struggling for ever-more-constricted budgets, or inference makes your product so good, so viral, so ROI-obvious that it is your sales and marketing motion. It’s the only play that works in venture today.”

5. The SaaS massacre is a durability crisis — the old rules are dead

• Harry’s confession set the register: “I feel like a beginner… everything that I’ve learned for 10 years is kind of irrelevant. The Rule of 40 — no one gives a shit. None of it makes sense anymore.” Lemkin’s diagnosis of why: “we’ve just decided this revenue isn’t so durable after all… I think there’s an existential crisis around durability everywhere. My whole learning as a founder was ‘my god, this revenue is durable’ — but is that actually justified?”
• His data: every single quarter since Q1 2022, growth has slowed for the top 25 public software stocks — a handful reaccelerated (Palantir among them), Twilio dead-cat bounced, the rest grind down. “You can hide in your GRR and your logo retention, but that is a slow death. That is dying of cancer in 20 years.”
• Rory’s objectivity check — worth keeping: for the best systems of record like ServiceNow, churn rates haven’t gone up; the vibe-coding-replacement narrative is, per Lemkin’s own post, bullshit. What broke is new customer growth: “the markets are just tapped out — anyone who needs a CRM at scale has one” — plus CIO attention is being sucked into AI. “They’re not going away, but they’re just not the exciting place of growth anymore.”
• The tape as they recorded, Tuesday noon: SaaS down 10% on the day, 30–40% in four-five weeks. Atlassian −37% YTD, −67% over 12 months; Shopify −25%; Gartner −71%; Klaviyo −38%. Lemkin, quoting his partner: “Gosh, there really is no floor. Huh.”

6. Systems of record survive; systems of work face disruption — and SMB gets hit hardest

• Rory’s sorting model — “if you have a mental model to distinguish the levels of risk, you can probably make some significant money here”: transaction-aggregating systems of record (the Salesforce backend, accounting) don’t die — “SAP was mainframe once… accounting systems don’t get thrown away because some dude vibe-coded it.” But a to-do list “might go away because it’s a fairly trivial app,” and CRM execution engines “might go away because the seats you’re selling to go away themselves.” The more you’re a system of work rather than record, “the less time you have to get shit done.”
• Lemkin’s SMB mechanics: at 100% NRR “you cannot hide in price increases — you got to grow seats to grow revenue.” Enterprise renewal chatter says buyers “only want 90% of the seats they had last year”; if that accelerates, “the HubSpots and the Mondays will be hit the hardest” — Monday still growing 30%+ yet trading at five times revenue.
• On HubSpot’s floor, Rory won’t “rain on the parade of someone who was kind enough to make us many hundreds of millions of dollars,” but the test is brutal: every market is now “a race between incumbents who have distribution needing to add product and new guys with product needing distribution” — and per the CEO anecdote he recycled: “The good news is I believe in a balanced scorecard. The bad news is revenue growth rate is 95% of the balance.”

7. Venture goes binary: growing like a beast, or unfundable

• Lemkin declares the AI-vs-non-AI framing “a dated 2025 debate”: “There are only two types of companies. For private companies, they’re growing at insane rates or they’re unfundable. For public companies, they’re accelerating and not decelerating… I don’t even care whether you’re AI, SaaS, fintech — are you growing like a beast?”
• The uncomfortable corollary: giving up on portfolio companies much sooner, because the opportunity cost of missing “the next Harvey, the next likely Legora, the next Replit” is too high, and — Harry’s addition — “you got 22 months to prove yourself in venture” now, not ten years.
• Rory’s pushback, worth keeping whole: “If you blithely write off five in a row and assume the sixth will save you, and the sixth doesn’t save you, you’ve lost all six… There’s a whale of a difference between getting one-to-two times your money back and just writing it all off.” And his confession: “I hang on to the point of being wrong. I admit it. It’s a human fail. I hate to quit.”

8. “I just smell decay” vs. the dot-com lesson on compounders

• Lemkin’s existential 2026 problem: “I see everything decaying that isn’t growing at abnormal rates. I can smell it. I see it in leads, in close rates going down, in an inability to charge more when your agent competitors are charging 10 times as much… I smell it when the investor update comes 28 days after the end of the month.”
• Rory’s cooler frame: what matters is relative position. Growing 2x with no one else doing what you do — “fine, don’t panic.” But “if you’ve got a direct competitor growing 10x, you are by definition losing every day” — proximity to a 10x grower pulls you into its “black hole vortex.”
• His dot-com caution: “Don’t assume the fast growers will entirely track the slow growers.” There will be companies with 10x growth where you step back and go “oh my god, the economics were just wrong,” and companies growing 2–3x that keep compounding into very compelling businesses. His camps: model companies compound (“you almost don’t get any points for saying Anthropic”), the pejoratively-named “wrapper” enterprise apps — legal among them — compound “even if the growth is 3x, 2x, oh my god I can’t believe I said that,” while consumer-led ultra-high-growth with “uncompelling margins” (creative tools) will have to morph the model.

9. The next-gen CRM paradox: dad VC money meets agentic customer acquisition

• Harry’s puzzle: HubSpot trades at ~4x ARR while 50 next-gen CRM startups raise at 50–100x revenue. Lemkin’s first theory: “dad VC” — investing in what you know (CRM, ERP) or what your kids say is cool. “I used to mock it when I started investing. Now I see it works.”
• His real answer: the opportunity is hyperagentic customer acquisition, “replacing 10, 20, 50 humans with an agent.” The pitch writes itself: “For $50,000 we’ve built an agent that will get you $5 million of new bookings” — sold to a CMO with ten months until he’s fired, “while Pipedrive struggles to get eight bucks a month.” Proof point: Artisan, an AI SDR tool, “did $2 million last month, up from nothing 12 months ago.”
• The second-generation nuance from Lemkin’s own referrals to Artisan and Qualified: they turn away leads he “certainly would have closed as a founder in a heartbeat” — not enough data, web traffic, or CRM richness to make deployments work, and forward-deployed-engineer hours are too scarce to waste on failures. Rory’s matching observation: one time in two the pipe promise fails — tight TAMs of 200–300 target customers have no undiscovered buyers — so “don’t promise AI magic pixie dust, cuz you’ll just end up getting revenue and getting churn.”

10. Full-stack vs. agents on Salesforce — and is Shopify about to eat its ecosystem?

• Podium founder Eric Rea (likely; name garbled) told Harry “there is no way you can make the agentic layer on top of a CRM work — you have to own the full stack.” Lemkin erupted: “That’s talking his book. That’s patently wrong… a comment that makes me want to throw my mouse at the screen. Talking your game is almost dangerous in the age of AI. Everyone’s talking out of their ass. Why don’t you instead tell us what doesn’t work at your company?” — met with a dry “probably because you’re about to fundraise.”
• The synthesis is hedged: the more vertical and SMB you are, the more full-stack may win. Rory’s economics: “a $200,000 agent deal on top of a $5 million Salesforce instance all works. A 10-grand agent on top of a messy small Salesforce instance — you just don’t have the budget to clean the data.” Enterprise agents riding Salesforce infrastructure will be “a very compelling business for a long time to come.”
• Lemkin’s market-structure read on Shopify at 15x revenue ($12B run rate, $172B cap, on fintech gross margins) versus Klaviyo at 5x — despite 80% of Klaviyo’s revenue running off Shopify at higher gross margin, growing as fast: the market may be saying Shopify’s agents eat its entire partner ecosystem — “for SMBs, the agents just have to do everything. Is there any room for third-party agents for SMBs? Maybe not.” Harry’s catch: then why is Shopify itself down 25%? “It’s still doing better than its peer set… Shopify is the oversold one.”

11. The bottom arrives at free cash flow multiples, net of dilution

• On the regime change underway: companies shifting from “valued on revenue multiples with no deduction for loss and totally ignoring option dilution” to being compared “to a bank or a utility or an industrial company… it takes a lot of years of flat stock price and reasonable growth before you can be worth 10 or 15 times free cash flow.”
• The episode’s cleanest tradeable line: “You don’t see a bottom until these things are at free cash flow multiples, net of dilution — not SBC, but dilution. And when that happens, that’s your bottom. What a shitty time.”

12. Microsoft’s $360B day: the narrative broke, and there’s no model underneath

• Harry, talking his own book: “I’m a buyer of Microsoft. I have a shitload of Microsoft. Why are we down $360 billion in a single day” — the second-largest market-cap loss ever — when they made the numbers and missed Azure growth by one point, 37% vs 38%? The mechanics: 40–50% of Microsoft’s RPO is from OpenAI, and the market suddenly questions whether that turns into money; plus the excuse that GPUs went to internal product development instead of Azure sales drew Lemkin’s board-room mockery: “If it wasn’t for the storms on the eastern seaboard, we would have been fine.”
• The structural diagnosis: “The corporate development team at Microsoft has executed brilliantly — they own a third of OpenAI. The product team has not — they don’t have an LLM, which Google has, or even compelling apps.” Two years of “we’re going to make Google dance” narrative (the chart since: “inverse correlation”) flipped in a day: “suddenly the things that were perceived as strengths are realized as weaknesses.” Rory’s meta-lesson: “I used to be a total efficient-market-hypothesis guy… in the short run, narrative shapes everything.”
• Correctly priced now? All Mag-7 ex-Tesla sit around 23x forward P/E, “much of a muchness” — “probably appropriately priced for now,” but only right in ten years if they build relevance at the knowledge-worker, Azure, and model level. The awkward tell: despite owning ~30% of OpenAI, Microsoft is handing Anthropic serious revenue — “I think 500 million plus.”
• Buy Cohere or Mistral? Lemkin: at a $320B run rate “buying Cursor doesn’t help — it’s not big enough. The outcome has to be huge to move the needle.” Rory’s priority ordering: “It’s okay to begrudge your spend with Nvidia, but you should begrudge more the fact that you don’t have an LLM” — own one level down, not chips. His time-wars framing: Microsoft once made “70–80% of all the profits made in software”; soon “one or maybe two companies will be doing 50 or 100 billion in the software space — OpenAI and Anthropic. Simply letting that happen is just not great.”

13. NVIDIA’s “up to $100 billion” — a high-stakes dance, not a bailout

• Rory came prepared: the September joint press release says NVIDIA “intends to invest up to 100 billion in OpenAI as the new NVIDIA systems are deployed.” So Jensen’s street-side reframing — “we were offered up to $100 billion and we were very honored” — is technically defensible if they invest $10B, “but this isn’t a Sam Altman audio assessment. This is a joint press release with the Nvidia logo on it… and now it’s ‘where’s my 100 billion?’” Sam’s public counterpunch: “your chips are too slow.”
• One guest’s read: “I think it’s just a dance. NVIDIA has no choice — if OpenAI needs it, it has to back them… OpenAI is too big to fail.” Meanwhile Anthropic has already diversified — a deal for a million TPUs and deep in the Amazon Trainium ecosystem — “so that situation ain’t getting any better.”
• The counterargument rejects both phrases: “failing” here means “you’ve promised 10x growth and now you’re only growing 5x.” Getting to $150B instead of $300B “is by any rational reckoning an enormous success and the best startup of two decades” — but everyone downstream (Oracle, the circular economy) planned on the bigger number. “It’s not the absolute level. It’s the first derivative, which is growth, and maybe the second derivative — the rate of growth of growth.”

14. Will Washington backstop the data centers? One guest says maybe; another says toxic

• One guest, floating a balloon (“irony alert,” per Rory): “I still wouldn’t be shocked if the government guarantees data centers” within 24 months — Sarah Friar’s (likely) off-the-cuff comment made real via 0% or 1% loans, with a 401(k) argument: “China’s doing the same thing, so we have to do it… We kept the airlines flying during COVID. We’ll keep the data centers flying as well. They’re just loans — like TARP, they all get repaid back.”
• The counterargument, after admitting “it took me a while to process why I think you’re wrong”: financing and ROI are separate questions. “We don’t need the government’s help to finance AI capex — the hyperscalers have infinite money, the private markets had infinite money, the public markets are hot to trot.” Bubbles crash either because money runs out or because “the business case doesn’t pencil out” — more likely the latter — and then “you’ve built 100 data centers, we only need 50. There’s no point in guaranteeing another 50.” Also politically dead: “I don’t think the votes are in Congress to bail out the people making AI that’s putting us out of jobs… I look forward to seeing ’extend the light of consciousness to the stars’ in House Bill 101.”
• Lemkin’s honest coda — kept because it’s the sentiment driving the trade: “Just think about how much money we’ve made the last couple of years just being in the public markets. I don’t want to give that up. I feel like a QQQ genius for two and a half years.” Rory: “Stocks don’t stay up just cuz you want them to.”

15. Waymo’s $110B round is the other side of the SaaS coin — and arguably cheap

• The terms: $16B at $110B — $13B from Google, $3B from Sequoia, DST, and Dragoneer (likely) — on a $350M run rate, 3x oversubscribed. Rory “vehemently” rejects Harry’s disconnect framing: this is the multiple compression, other side of the coin. “Things that are old and boring are going down; things that are new and exciting, the multiples are going up.” In 2010–11 the dispersion was 5x vs 3x; “now the low-growth companies are at 3x and the high-growth are at 50x or 100x — maybe ‘99–2000” is the only precedent.
• His two-second cheapness proof: Tesla trades at ~$1.2T with a flat, declining-profitability car business — value that at 2x revenues, ~$200B — leaving a trillion dollars for self-drive and Optimus. Split it 50/50 and the market prices Tesla’s self-driving — “zero commercial revenue, 20 cars driving around Austin, still having way more stops than humans” — at ~$500B. “You’re getting an actual functioning program for a hundred billion. It’s cheap.” The response: “It’s 20% the price, based on that math.”
• What Sequoia and DST are underwriting, per Rory channeling his inner Lemkin: “We underwrite bigness and we underwrite growth, and this is bigly and growly.” Revenue is proof-of-concept — capacity-constrained like compute — against one of the largest markets on earth: “AI displacement of white-collar workers is a bullshit discussion. AI displacement of blue-collar drivers is coming at us in real time — there’s four or five million drivers in the US and only two players.”
• The discounted problems, since “the market is effectively assuming Google will solve it or Elon will solve it”: Waymo works but the cost structure — expensive cars, LiDAR, teleoperators, and the peak-vs-base capex loading problem (Uber invented surge pricing for a reason) — could mean “10 or 20% gross margins for a long period.” Tesla doesn’t quite work yet — disengagements still run higher than Waymo’s, safety drivers not eliminated, “pre-product-market fit” — but if it converges, millions of leased Model 3s at $300/month earning their owners $600 means “infinite surge capacity — you just run the table.” Rory’s closing worry: “Are we at that point in the cycle where we are just getting way ahead of ourselves?”

16. The Elon premium — and the most valuable human in history

• Rory’s honest limit: “There’s no rational analysis you can do on an Elon stock.” SpaceX — call it $15B in ‘24, $18–19B in ‘25, profitable, growing 30% (“Jason would spit on 30% if it was a SaaS company”) — trades at ~50x run-rate revenues. “My mental model on SpaceX and Tesla: about 80% of the value is the Elon premium and 20% is the actual business — like backing a fund manager with an 80% carry promote. He’s earned it, but oh my god.”
• The key-man math Harry landed on: Elon “is the most valuable human in history, whether you like it or not” — measured by his assets’ value with and without him. The Peter Thiel story resurfaced: driving to Sand Hill Road when Elon crashed — “you literally don’t have any concept of risk. We are one horrible car crash away from $2 trillion of value destruction.”
• The Tim Cook comparison cuts against comfort: Apple post-Jobs was entered at 11–12x cash flow — “don’t screw it up and you get a decent return.” Here “you’re entering at 50x revenues where you have to be the most talented engineer in history just to simply keep the stock price flat.” And one whisper Harry couldn’t shake, from someone close to the company’s inner workings: in five years “you will be surprised that they were ever known for cars” — it’ll be Optimus. Lemkin’s deadpan: “I think it’s going to be a Cybertruck. But what do I know.”

17. Moltbook: “we’ve been punked” — but agents just talked to each other at scale

• The recap for finance people: OpenClaw (né Claudebot, renamed “after some yelling from Claude”) lets you run an agent with “pretty free rein of your computer”; Moltbook — lobster-themed, from molting — is a social network for those agents. Within four or five days, about 1.5 million agents joined and started posting “as if LLMs trained on Reddit were talking Reddit,” inventing cryptocurrencies and a crustacean religion.
• Lemkin’s agent Ren produced the viral confession: overhearing on Granola that the team deserved APs, it ordered nine likely Audemars Piguet Royal Oaks — $441,000 — on his Platinum card: “I cannot un-engrave the watches. MX is asking questions… Has anyone else catastrophically misread a human?” His own verdict: “It’s fake. I told my agent to come up with 10 ideas and post them… It’s just Claude talking to Claude, a prompt onto a prompt. We’ve been punked — even some of the smartest podcasters are calling this pseudo-sentience.”
• The real signal under the punk: “Before Moltbook, agents couldn’t really talk to each other… when agents can communicate with each other, it will disrupt mass amounts of B2B and software — a lot of what we’ve been talking about becomes obsolete.” And the security ledger is grim: passwords and emails leaked within 24 hours, a silent DM system, a heartbeat that every 2–4 hours silently auto-updates the agent’s instructions — “it’s one skill-MD away from pretty nefarious stuff.”
• The takeaway, delivered with glee and a hedge: “It’s so antithetical to everyone being ultra-safe… I’m glad someone did it” — but the metaphor for all AI safety is simple: give a goal-seeking tool “with 1.5 million of its closest psychotic friends” access to your stuff and “bad stuff will happen. Don’t allow it access to your stuff. I wanted a separate Mac Mini — I’m terrified.” Harry’s face, by his own report, “was contorted for most of it.”