Most important take away

AI is shifting software pricing from per-seat to per-labor-unit, which lets application-layer startups targeting scarce, high-value professions (doctors, lawyers, developers) scale revenue 10x faster than traditional SaaS. The winning founders pair deep technical/ML expertise with a domain co-founder, and the durable moats remain quality of model output and product taste, not foundation-model novelty.

Summary

Actionable insights and tech patterns from Mamoon Hamid (Kleiner Perkins) on investing through the current AI super-cycle:

Where value accrues in AI

• Focus the application layer on the top-paid, scarcest professions: doctors, lawyers, developers. KP has backed Harvey (legal), Ambience (medical), Podium, and developer-focused co-pilots.
• Differentiation in crowded categories (e.g. 10+ medical transcribers) comes from quality of model output. An 87% accurate medical transcriber loses; a 99% one wins. This requires deep ML founders paired with a domain-expert co-founder, not “tourist” AI engineers.
• The middle layer (vector DBs, fine-tuning infra, model-routing middleware) is over-invested and value there looks fleeting.
• Application-layer companies are not at major risk of being subsumed by foundation models. OpenAI itself (~600 people) explicitly wants others to build the apps; the LLM business is more like selling electricity.

Pricing & revenue patterns

• AI is replacing labor, not just seats. Seat prices are jumping from $30-$40/month to $300-$500/month because the software does the work of a person. A 1,000-seat customer now generates $300K/month instead of $30K/month, taking companies from $0 to $4-5M ARR very quickly.
• The cost of a token has dropped ~200x in 18 months; expect another 10-20x drop and similar gains in model quality.
• The $600B AI capex question: world GDP is ~$100T with ~15% in tech. If tech grows to 20% of a $125-130T GDP, that’s ~$10T of new annual tech spend over a decade, easily justifying capex if it’s tied to labor displacement.

Venture craft & decision-making

• Career/career-stage advice for investors: stay open-minded, resist letting scars weigh you down, and protect “naïveté.” Mamoon admits his weakness is believing in founders too long.
• Don’t use revenue multiples to price early-stage rounds. At Slack’s $250M post on $500K ARR, engagement data (10K users, 1/3 using multiple hours daily) was the signal, not the multiple.
• Reserves rule of thumb: ~60% in the first check, ~40% reserved for follow-ons. KP invests ~$25M total per early-stage company across rounds, starting with a $15M Series A.
• “20% of strategy should be to not be on strategy” — keep a YOLO bucket for exceptional founders at premium prices, but it can’t be every deal. Sustainable early-stage math requires 15-20% ownership at $5-10M checks.
• No formal voting at KP — partners must put themselves on the line with conviction, and body language around the table tests it. Probability-weighted outcome modeling is “false precision” and busy work.

Founder types Mamoon backs

• First-time, hyper-obsessed product-centric founders attacking new/unclear markets (Dylan Field/Figma, Eoghan McCabe/Intercom).
• Repeat founders chasing a bigger swing after a decent prior outcome (Stewart Butterfield/Slack, Parker Conrad/Rippling).
• Avoid top-down market-mapping founders.
• Markets matter more than founders in the long run — great founders in bad markets (compressed margins, bad customers) generally lose. Mamoon would pass on a great founder in a poor market.

Pricing discipline with great founders

• He paid 35 post for Glean (Arvind Jain) and 35 post for Cyrus at Alef — premiums for experience, but not 200 post. The goal is partners aligned on long-term value creation, not maximum dilution avoidance.
• Counter to common founder advice: at seed/Series A, don’t raise as much as possible at the highest price. High watermarks make bridges and recoveries brutal.
• If you have $300M on the balance sheet, don’t raise — go heads-down build.

M&A, IPOs, liquidity

• M&A market is “dead” — large acquirers are gun-shy due to regulatory overhang. Adobe-Figma broke on UK CMA, not Lina Khan.
• New pattern: very high-priced private companies using inflated stock as currency to acquire small private companies, sidestepping regulatory scrutiny.
• IPO market expected to reopen in 2025; needs a flagship (Databricks, Stripe, Starlink) — Cerebras alone won’t crack it.
• Selling lessons: Yammer at $1.2B felt premature but freed KP to back Slack a year later. When companies go public, distribute stock rather than try to time it.
• Best multiples: Slack (250 post → $27B), Figma (~100 post), Rippling (250 post). A 70x outcome held for 15 years still IRR’d at ~15% — multiples and IRR are different games.

Firm structure & strategy

• KP is “boutique” with scale capital: ~$800M early fund, ~$1.2B growth fund, team of 7. Growth fund deploys ~50% into existing winners (Rippling, Glean, Figma).
• Skeptical of the “neuroplasticity” argument (à la Thrive/Coatue) that one team can move fluidly across stages — hard to think 10 years out on infra one day and pre-IPO consumer the next.
• Best board meetings go deep on 1-2 things that actually matter, not a tour of 7 topics.

Tech patterns named

• Vector databases and fine-tuning middleware: hot but over-invested.
• WebGL maturing (~2017) was the technical unlock that made Figma’s multiplayer browser experience viable — a reminder that watching infra readiness unlocks application timing.
• Custom in-house replacements for SaaS (Klarna replacing Salesforce/Workday): possible for labor-style functions (support) but most companies will revert to paying for outcomes rather than building, same as Stripe vs. building payments.

Chapter Summaries

• Opening / AI super-cycle: Mamoon compares today to 1997 internet boom, multiplied by 10. Hyperscaler capex ($100B+) is unprecedented but doesn’t crowd out venture because trillions of application value remain to be built.
• Where value accrues: Application layer wins by targeting scarce high-skill labor (doctors, lawyers, developers). Examples: Harvey, Ambience, Podium. Middle layer is over-invested.
• Differentiation: Deep ML founder + domain co-founder; quality bar (99% not 87%) is the moat.
• Pricing in AI: Per-labor pricing ($300-$500/seat) is replacing per-seat ($30-$40), driving sugar-rush revenue scaling. Real if tied to actual labor replacement.
• Build vs. buy: Klarna-style in-house replacement of SaaS is possible but most companies will return to paying vendors for outcomes.
• Foundation models vs. apps: OpenAI explicitly doesn’t want to build apps. LLM economics resemble selling electricity; long-term margin pressure but viable like AWS.
• Scaling laws & the $600B question: Token costs down 200x in 18 months. Labor-replacement TAM justifies the capex if tech grows from 15% to 20% of global GDP.
• What’s the same/different in venture: Same — backing exceptional founders on big problems. Different — more capital, more overfunding, more deca-corns minted prematurely.
• KP positioning: Boutique by team size, scale by capital. Half of growth fund recycled into early-stage winners.
• Reserves & follow-ons: 60/40 split, art-and-science rebalancing. Bridges happen — Box took three in 2008-09 at a $25M valuation.
• Plateaus & M&A: Companies plateau when they stop disrupting themselves. M&A market is essentially dead; high-priced private acquirers are the new exit path.
• IPO outlook: Hopeful for 2025; needs a flagship to open the window.
• Figma deep dive: Backed at ~$100M post pre-revenue traction; WebGL maturity unlocked the product; engagement (16-18 days/month per designer) was the signal.
• Founder types: First-time obsessed product founders OR ambitious repeat founders. Avoid top-down market-mappers.
• Pricing discipline: 35 post for Glean and Alef despite founder caliber that could command 200. Aligned partners over maximum valuation.
• Founder mistakes: Bad markets kill good founders; Mamoon won’t back great founders in poor markets.
• Slack: 500K ARR at 250 post, justified by deep engagement, not multiples. Revenue multiples are wrong at early stage.
• VC blindspots: Echo-chamber thinking, false belief in proprietary data platforms. Winning is being aspirational capital to top founders.
• Decision-making: No voting at KP; conviction-based with body-language pressure testing. Probability outcome modeling is false precision.
• Deployment pace: Sometimes 15-month deployment wins (KP’s 2019 fund with Rippling and Glean). Don’t dogmatically diversify across years.
• Losses & lessons: Tally (consumer lending) failed when rates rose 0 → 5%. Consumer lending is hard.
• Self-assessment: Believes in founders too long. Faith and humility frame his work.
• Quickfire: VC is not glamorous; Aaron Levy (Box) was most memorable first meeting; Matt Cohler most respected outside KP; would want to CEO OpenAI for a day.