Jun 4, 2026 · 1h 38m · news
Anthropic Files to Go Public | Cognition Raises $1BN at $26BN Valuation | The 996 Work Ethic · 20VC with Harry Stebbings
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of 20VC, host Harry Stebbings, Jason Lemkin, and Rory O'Driscoll debate major industry developments including Anthropic's confidential IPO filing, Cognition's massive valuation, the shift from traditional SaaS models to token-based AI agents, and the demanding work cultures of high-stakes tech startups.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 8.2% of the talking time here. How this is scored →
Harry as informed peer 3.9 Guest teaching 4.3 Guest disagreement 4.0 Harry pushing back 3.9
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 3:16 Rory rejects Jason's trillion-dollar startup focus
Rory forcefully rejects Jason's assertion that VCs shouldn't bother taking meetings unless a company can yield a $1B position, telling him to 'grow up and be a fucking adult' and manage standard fund distributions.
Hardest push from Harry ▶ 7:32 Harry challenges Jason's position-sizing logicHarry directly refuses Jason's claim that investors can filter out deals without $1B position potential upfront, citing evidence from 1,000 interviews with top GPs who consistently underestimated their largest winners.
Biggest teaching moment ▶ 1:21:29 Rory explains private equity SaaS debt structuresRory clearly educates the room on private equity buyout mechanics, demonstrating mathematically how 10x revenue acquisitions leveraged with 5x debt wipe out equity holders when multiples compress to 5x.
Harry holds his own ▶ 7:32 Harry leverages interview archive to dispute guest premiseHarry uses his empirical background interviewing over 1,000 venture capital fund managers to counter Jason's rule about identifying outcome sizes during early-stage screening.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Anthropic's IPO Filing and the Hunt for Trillion-Dollar Startups | 6 | 4 | 6 | 7 | Harry actively challenges Jason's claim that VCs should only target potential billion-dollar positions, drawing on insights from interviewing over 1,000 top fund managers who underwrote massive winners without predicting outcome sizes. Rory forcefully rejects Jason's framing as unrealistic, urging investors to accept normal distribution models instead of chasing outlier outcomes. Rory also offers a quick correction on his fund size when Harry misstates it as a billion dollars. | |
| Is the SaaS-pocalypse Over? | 3 | 4 | 3 | 2 | Harry sets up the discussion on public SaaS earnings surges, leaving the floor to the guests to break down market dynamics. Rory details his cloud ETF trades while Jason outlines the structural shift from seat-based licenses to token consumption. The guests maintain a collaborative and analytical dynamic with minimal host interruption. | |
| Cognition's $1B Valuation and Autonomous AI Engineers | 2 | 3 | 2 | 2 | Harry asks Jason to compare Cursor and Cognition valuations following Cognition's $1B raise. Jason provides context on CTO adoption of Devin and explains why autonomous AI engineers present a larger opportunity than copilots. Harry listens without intervening or challenging the premises. | |
| The Token Budget Crisis, Multiple Models, and AI Layoffs | 4 | 5 | 7 | 4 | Rory and Jason engage in a heated debate over whether companies will substitute human developers with token budgets by year-end. Rory scrutinizes Jason's math, arguing that replacing headcount with tokens requires far higher per-engineer budgets than current corporate reality. Harry brings in external examples like Salesforce spend to steer the discussion. | |
| AI and the Shifting Landscape of the Legal Industry | 4 | 5 | 4 | 3 | Harry introduces Kirkland & Ellis's $100M internal AI investment and the movement of legal tech CEOs to foundation model companies. Jason frames Kirkland's commitment as a modest PR move, while Rory breaks down why law firms protect proprietary workflows and data from third-party model training. Harry asks clarifying questions on full-stack legal services. | |
| The Potential of AI Wealth Managers | 5 | 5 | 4 | 6 | Harry pushes back on Jason's optimism regarding Robinhood AI investment agents, questioning whether automated trading leads to market-wide trade commoditization. Jason and Rory clarify the distinction between automated financial planning, which benefits users, and generating trading alpha, which remains unproven for LLMs. | |
| AI Agents as Domain Experts | 5 | 4 | 4 | 6 | Harry points out a logical contradiction in Jason's framing, asking how product design can simultaneously educate users into domain experts while automating execution on their behalf. Jason responds by delineating user education from agent autonomy. | |
| Private Equity Software Returns and LP Realities | 4 | 5 | 3 | 3 | Harry introduces Apollo's warning about private equity software returns and Harvard's high private asset allocation. Rory outlines the unit economics of leveraged buyouts, explaining how 10x purchase multiples with 5x debt leverage destroy equity value when growth slows and multiples compress to 5x. | |
| VC Windfalls, Retention, and Partner Motivations | 4 | 4 | 3 | 4 | Harry asks how venture firms handle massive carry distributions like Anthropic's without losing key partner motivation. Jason and Rory explain GP retention mechanics, mentioning how partners stay motivated or wind down firms based on capital commitments and personal drive. | |
| Startup Work Ethic and the 996 Debate | 3 | 4 | 4 | 3 | Harry prompts a discussion on intense startup work cultures like Corgi's 996 schedule. Jason argues that extreme work demands are acceptable if early employees receive life-changing equity upside, while Rory warns that excessive work intensity degrades strategic judgment. | |
| Burnout, the AI Automation Irony, and Tech Talent Realities | 3 | 4 | 4 | 3 | Rory highlights the irony between widespread claims of total AI work automation and the reality that tech workers are working harder than ever while struggling to recruit top talent. Harry agrees, pointing to his own hiring struggles. |
Statements from this episode (30)
O'Driscoll: Investors cannot understand losing money without experiencing it directly
“Losing money is like sex. You can talk about it all you like, but until you feel it, you don't know what it's like.”
O'Driscoll: Tech businesses shifted from cash generators to capex-heavy cash consumers
“All these businesses have gone from capex-like cash flow machines to capex-heavy cash consumptive machines.”
Calacanis: Developers would quit if denied their choice of AI model
“I would quit as a developer if you told me I could not use the model of my choice. I would quit.”
Calacanis: Companies will choose spending on AI tokens over humans by year-end
“I really do think by the end of the year, we're going to choose tokens over humans.”
O'Driscoll: Normal years won't yield over 5 exits above $10B
“Really there's not going to be more than four or five, ten billion dollar plus outcomes a year.”
Rory O'Driscoll: Cash-consumptive CapEx-heavy businesses underperform stock markets medium-term
“All these businesses have gone from CapEx light cash flow machines to CapEx heavy cash consumptive machines, right? Generally, that's never good for stock prices over the medium term. Just, oh, across history, things that have high cash flow spinning out are r…”
Rory O'Driscoll: AI data center investments typically require 2 to 3 year payback
“The payback on these AI data centers is normally two to three years. Now, Elon has massively outperformed that with his storage deal with Anthropic where he's getting all his money back if the deal lasts a year and a bit. But normally the payback is not, it's …”
O'Driscoll: Current frontier AI models will be 5x-10x cheaper in a year
“The frontier models aren't getting cheaper. I want to be clear on that. They're getting actually slightly more expensive, but the model that is frontier today will in a year be five to 10 X cheaper because it won't be the frontier model anymore.”
Calacanis: Most AI coding tokens are spent on QA rather than production
“The vast majority of tokens that are used in coding are in QA anyway. It's not in production.”
Calacanis: Replit builds features with Claude Sonnet and verifies code via Codex
“Replit, if you have a complex feature, Replit builds it in Claude, Sonnet, not Open, it builds it in Sonnet to save money, and then it has Codex come in and check the work. It has both of them.”
O'Driscoll: Enterprise AI spend has established a $500BN-$1TN market size
“What's happened is somewhere around three, five percent of tech spend. Everyone noticed, oh my God, we're spending this. And no one said, we're going to cut back to zero. What this means is you've just established a category that probably has a market size of …”
Calacanis: Software QA departments will be destroyed by 2027 due to AI
“I think what will happen going into 2027 is an engineering QA departments will get destroyed because you'll be like, I got six QA engineers.”
Rory: EDA software tools account for 13% of engineering budget per engineer
“The EDA software market, which is the most automated. Market today in terms of tools relative to engineering spend is roughly 13%. So for every engineer you hire, you allocate 13% for EDA.”
Calacanis: $100M Portfolio Company Tripling Product Count This Year
“And so I've got one company crossing a hundred million that literally has, was going to end this year with three times more products than it did last year.”
Calacanis: AI Startups Will Eventually Re-Bloat to Historic Headcounts
“And so even, and so our startups that we're excited about, I think ultimately they will achieve the same historic level of bloat, maybe half the size, but they will get as bloated as they can because they will have much larger, broader product lines.”
Rory O'Driscoll: AI Will Not Boost GDP Productivity Growth Above 2%
“Or where I stand, the statistic, the average over the last 200 years has been two percent. It'll stay at two percent.”
Rory O'Driscoll: Non-Engineering Bottlenecks Limit Overall Startup Speed
“The weakest link in the chain is what determines the speed of the convoy, the wagon trainer, in this case, the company.”
Calacanis: Kirkland & Ellis generates $11B revenue growing 20%
“It's a law firm that does eleven billion in revenue growing 20%.”
O'Driscoll: AI vendors will standardly sell software to law firms in 5 years
“Fast forward five years, to be clear, I think there will be AI-focused service providers, two law firms, they'll buy the product and the drama will be out of the deal.”
O'Driscoll: Full-stack AI law firms won't replace elite law firms
“I don't think the full stack law firms will replace K and E.”
O'Driscoll: LLMs are unproven at trading stocks and beating human pod managers
“The thing LLMs have been somewhat unproven as of yet is the ability to actually trade stocks. You basically be a pod manager in a hedge fund world and outperform humans. The record on that isn't there yet.”
Stebbings: Harvard endowment holds 41% of portfolio in private assets
“Harvard saying that now, 41% of their book is now privates.”
O'Driscoll: Overpaying for Decelerating SaaS Companies Leaves No Operational Escape
“In companies that are growing at 20% and then suddenly slow down to eight or nine percent growth rate, if you've overpaid, it's kind of like overpaying for a real estate transaction. There's nothing you can do. There's no accelerant. There's no magic that's go…”
Calacanis: Massive AI Distributions Will Mask Underperforming VC Funds
“I do think these distributions will, I think the distributions will facilitate us somewhat ignoring maybe some bad funds.”
Stebbings: Menlo to make $10B carry on Anthropic, Founders Fund more on SpaceX
“Menlo will make ten billion dollars in carry Spot will too, plus Founders Fund will make more than that from SpaceX.”
O'Driscoll: Large Wealth and Free Time Prove Destructive in 30s-50s
“Large amounts of money and large amounts of free time tends to be pretty destructive, especially, you know, thirties, forties, and fifties, you know, it's hard to fill your day with, and there's so many things to do wrong.”
O'Driscoll: Unpaid Failure Is Statistically the Most Common Startup Outcome
“I had my own startup. It didn't work out. I look back and I worked, you know, seven by 24 for three years and made no money. That sucked. Right. And that's as a reminder, that's the mode of experience in the correct sense of statistics. That's the most single …”
Rory O'Driscoll: Extreme work hours increase output but degrade decision-making judgment
“I find when you're working really intensely and you're stressed and you're kind of caught up in something, the good news is you put in an extra 10 hours of effort. The bad news is you lost your judgment in doing it.”
Rory O'Driscoll: AI will not cause mass white-collar unemployment in 3 years
“And we've grown to accept that, that, you know, we're all here in the Valley with a plan. To automate white collar work such that there's going to be mass unemployment, according to these folks who are totally wrong, in my opinion, in three years, and all the …”
Rory O'Driscoll: Tech workers will stay employed, but at lower compensation
“The truth is this, people will get jobs. They mightn't get the high step. I mean, it sucks, but the people repurpose from, maybe you won't get another job that pays 400 grand and allows you to work from home three years a week. I think you will get a job, righ…”