Most important take away

Foundation models are the fastest depreciating asset in history — the lasting value lives in talent, data moats, and applied workflows, not the models themselves. Venture is not an “asset class”: only a tiny number of companies actually generate liquidity, and most investors will lose money chasing AI logos, replicated geographies, and PE buyers who are not coming. If you have conviction and an information advantage, take the price the public markets (or buyers) offer now — ride winners, but don’t romanticize holding.

Summary

Actionable insights and career/tech patterns from the conversation:

Investing and capital strategy

• Hold two opposing truths at once: AI is the most transformational technology of our lifetime AND a financial bubble that will destroy capital. Both are simultaneously true, like the dot-com / fiber-optic era.
• Foundation models are the fastest depreciating asset in history (credit to Gavin Baker). Don’t invest in a business “where the assets walk out at night.” Value accrues to lasting moats (data, distribution, unique workflows), not the models.
• LPs frequently lack true diversification — many funds pile into the same hot companies. The “logo chasing” pattern (must be in OpenAI, Anthropic, Character, etc.) destroys returns.
• Don’t treat venture as an asset class. It is sui generis — a tiny number of companies and craftsmen actually make money. ~10 generational companies per year drive most returns.
• Premium multiples accrue to the only way to play a future trend in public markets (Nvidia is the prime example). Being unique has real value beyond growth and margins.

Where AI value lands

• Break the “AI bucket” apart: foundation models, applied AI, vertical workflows. Most foundation model bets will fail; one or two will return capital.
• “AI countries vs. non-AI countries” will define the 21st century the way nuclear did the 20th. Israel and US are advantaged; Europe will fall further behind because of over-regulation — if you’re starting a company in Europe, “get out.”
• Horizontal SaaS is at peak — pulling-forward-the-model thinking is over. Vertical SaaS may be eaten by AI because customization gets cheap. Accenture and McKinsey are quietly winning AI revenue ($2.4B generative AI at Accenture) because companies don’t know how to integrate AI themselves.
• “Amazon told its teams: don’t buy more software, AI will handle business process needs” — possibly apocryphal but a real directional signal.
• Buyers pay for value/outcomes now, not seats. Price-per-seat SaaS pricing must adapt to “people pay for work.”

Career and operating patterns

• Avoid competitive markets — high CAC, low retention, price erosion, brutal product marketing. Prefer non-consensus right (Howard Marks): uncharted spaces where you have unique domain knowledge.
• Patriotism in tech is legitimate, but “margin is my opportunity” framing of defense is wrong. Defense sales are a craft (Palantir took 20 years to crack the code) — MBA frameworks lose money there.
• Hard-tech founders increasingly won’t take SaaS-trained capital because the SaaS playbook doesn’t apply. If you’re a software investor moving to deep tech, expect Darwinian losses.
• The IPO window is open at the right price. Take companies public early (Lemonade IPO’d at $60M revenue, Shopify at $700M). Don’t wait for $500M revenue and a top-tier banker — raise $80M off a $600M IPO and build efficiently in public.
• PE is not the savior of M&A. Higher rates compress multiples; many PE-bought SaaS assets are underwater (Pluralsite write-down is just the start). Lina Khan blocking M&A is “a threat to American capitalism” — she needs to go regardless of administration.
• 98% of investments don’t produce liquidity. Get real about DPI vs. TVPI — though TVPI still matters if the underlying compounding is durable. The deeper question: how durable is the moat producing those future cash flows?

Portfolio construction discipline

• Ride winners, but if you have an information advantage and conviction the price is too high, sell 100%. Don’t average in or out — venture is binary. Eisenberg’s biggest mistake: not selling 100% of one or two positions at peak.
• Don’t do pro rata reflexively. Buy ownership in the first round when your insight is sharpest. Pro rata into “middling” companies sometimes pays off (Jesse Beyroutey insight) — winners run away from you, losers don’t deserve it.
• “Spray and pray then concentrate into winners” is a lie — by the time winners are clear, they’re oversubscribed.
• Consumer is fast fashion (ephemeral); enterprise is more predictable. Sell consumer winners aggressively on the way up.

Board meetings and EQ

• Most board meetings can be 45 minutes with proper prep: a letter (not deck) rich in data, 1–3 strategic questions surfaced upfront, calls beforehand. Three-hour board meetings are pontification.
• In-person beats Zoom for relationships and reduces grandstanding.
• Don’t be the product manager from the board seat. Don’t get too absorbed — first-time founders get rattled by intensity.
• Prefer naive founders in heavily-regulated industries (Lemonade founders knew nothing about insurance — that was the advantage). Healthy skepticism of expertise.

Personal and macro

• Younger people will surprise us — the cohort that has had to stand up and fight in real conflicts has a maturity MBAs lack.
• China is likely ahead, not behind — assume your competitor is 2x better than you and work accordingly.
• China investing was a ZIRP phenomenon — “stroke of pen risk” (government seizure) was systematically ignored in the yield chase.
• The biggest ZIRP sin: when things are too easy, they don’t matter. Hard things — relationships, defending freedom, building real companies — are what produce satisfaction.
• There is no work-life balance. The harder question is how the people you love perceive how you spend your time.

Chapter Summaries

• The AI bubble paradox: AI is the most transformational tech of his lifetime AND a financial gold rush that will destroy capital. Dot-com and fiber-optic parallels.
• Who loses money: LPs with hidden concentration across funds, logo-chasers, geographic copycats, and foundation model investors who don’t own a durable moat.
• Foundation models depreciate fastest of any asset; value accrues to talent and applied layers. Skepticism on Perplexity’s “value is in the team” framing — humans walk out at night.
• Where AI value lands: hyperscalers acquire smaller foundation models; Anthropic vs. OpenAI fight at the top; the real prize is owning where users start their day.
• Venture is not an asset class. A tiny number of companies make all the returns.
• Breaking down the “AI bucket”: foundation models (impossible from Israel), applied AI, full-stack vertical AI. AI countries vs. non-AI countries will define the century.
• Why competitive markets are traps; premium multiples go to the unique way to play a trend (Nvidia thesis).
• Peak SaaS: horizontal SaaS over-saturated, vertical SaaS at risk from AI. Accenture/McKinsey winning because companies can’t self-integrate AI.
• Pricing for value vs. seats; adoption cycles are slower than hype, faster than ever in capability.
• Regulation as the real plateau risk — Europe especially. Defense, kill chains, and the China-AI race; respect your enemy as 2x better.
• The defense gold rush attracts MBAs who don’t understand the craft (Palantir took 20 years; Anduril broke through because of Palantir + Trae Stephens).
• Liquidity reality check: M&A blocked by Lina Khan, PE constrained by higher rates and underwater SaaS portfolios, IPO window open at lower prices. Go public early.
• Secondary markets are mostly fake — buyers are data tourists doing diligence under the guise of LOIs.
• Portfolio construction: ride winners but sell 100% when conviction warrants; binary investments, no averaging. Pro rata into middlers sometimes wins.
• TVPI vs. DPI debate: DPI matters but TVPI is real if the underlying moat is real. Spend more time analyzing competitive advantage.
• The middle-growth trap (10–20% growth) — these would have been great businesses without venture capital on the cap table; secondary trades and buybacks are the exits.
• Self-critique: too direct, too cranky in board meetings, too absorbed in companies. Prefers naive founders in regulated industries (Lemonade).
• Quick-fire: changed mind on younger people; biggest misconception about Israel is scale; respects all competitors as 2x better; biggest ZIRP sin was not selling more and China investing; aspires to remain relevant past 53 in a young-person’s business.