The lean-ops thesis got five more voices, and the same operating model has a name now: AI founder mode

The Threads

Five voices, one operating model: AI founder mode has a name now

Last week the digest’s central observation was the lean-ops pattern emerging from three independent verticals (AppLovin, Kalshi, Baseten). This week the same pattern got named, by Brian Chesky — AI founder mode — and five additional voices independently said the same five things:

1. ‘People-managers-only do not survive the AI age.’ Chesky direct quote; Max Schoening (Notion) independently the same sentence; Toby Lütke (‘every engineer needs to code; even the managers need to code’); Adam Foroughi (‘Anyone on LinkedIn that talks about their team should be fired. Lead from the front with AI’); Brian Armstrong (Coinbase’s 14% layoffs and the ‘I just don’t need anyone at Coinbase that isn’t an individual contributor anymore’ edict). Five independent voices, one week.

2. ‘There will be no apps in the future — only agents.’ Direct verbatim quote from both Chesky and Schoening this week, independently. When two product leaders at $100B+ companies say the same line in the same week, it’s worth treating as the architectural call to plan around.

3. The lean-ops financial outcome is now visible. AppLovin: 84% EBITDA margins, ~$10M EBITDA per employee in the 400-person core. Shopify: 7,500-8,000 people targeted to stay flat for 5 more years ‘at 100x productivity per head.’ Airbnb: Project Hawaii — 10-12 person Navy-SEAL teams — produced $200M of internal revenue in year 1, $400-500M in year 2, ~600 basis points improvement on $134B GMV by year 3. The pattern is now a methodology that ships, not just a hiring filter.

4. Senior engineers + steering beats juniors + no-priors for AI coding. Toby Lütke explicitly changed his mind on this — initially believed AI-native juniors would have the advantage; now: ‘all engineers are massively underestimating how important the steering is.’ Foroughi: ‘10x engineer × AI = 100x. B player × AI = 2x. The gap widens, not closes.’ Direct contradiction of the early-2025 ‘AI flattens the experience curve’ framing.

5. Context engineering as the new dominant role. Toby calls it ‘context engineering will become a recognisable career role that subsumes more specialised roles.’ Schoening calls it the same thing from the Notion side. Cross-references the Issue 02 PM-debate resolution and Foroughi’s ‘engineers as PMs’ framing. The PM role isn’t dying — it’s morphing into the highest-leverage role if the person has actual technical and product taste.

The most concrete production-deployed example came from Toby Lütke: River — Shopify’s agentic engineering system that lives in Slack, self-named, does ‘a ludicrous amount’ of Shopify engineering, with engineers steering it via public Slack channels so the whole org learns from each other. Combined with Cat Wu’s Anthropic engineering velocity disclosures from Issue 02 (‘6 months → 1 month → 1 day’), this is the operator-side proof that the architectural shift is happening at the production layer of multi-$100B companies.

Where I’d put numbers on this:

• At least 3 more publicly-traded software companies (excluding AppLovin) post EBITDA-per-employee greater than $5M in their 2026 annual disclosures [forecast: 2026-05-10-001] — Foroughi-style operating-leverage profiles are about to become a cohort, not a one-off. Confidence 0.55.
• ‘AI founder mode’ becomes a standard analyst / sell-side framing within 6 months (cited in at least 3 sell-side equity research reports or earnings-call CEO comments) [forecast: 2026-05-10-002] — Chesky just named it; the language travels. Confidence 0.60.
• Net headcount across the top-20 US technology companies declines by ≥3% YoY by Q4 2026 [forecast: 2026-05-10-003] — even with Toby’s counter-narrative that ‘most layoffs are COVID-era overhiring,’ the destination is the same. Confidence 0.55.

Anthropic: ‘biggest monopoly in human history’ — or 4-horse race?

The single highest-stakes framing of the issue. David Sacks on All-In: Anthropic ARR went $10B (Jan 1) → $30B (Mar 31) → $44B (April 2026). If 10x/year exponential holds for 18 more months: $1T ARR by 2027 — bigger than the rest of Mag 7 combined. Sacks calls this ‘the biggest monopoly in human history’ and reverse-engineers the Rockefeller-rebrand: ‘Imagine if Rockefeller had called his company Safe Oil and asked Washington for a new agency to regulate kerosene safety. People would have debated wick thickness while he built the biggest monopoly in history.’

The supply side moved equally fast. Chamath’s prediction from the prior week landed in 5 days: Elon’s xAI leased all of Colossus 1 — 220,000+ Nvidia GPUs and 300+ MW — to Anthropic. xAI shifts training to Colossus 2. Claude Code rate limits doubled. Peak caps removed for paid users. Anthropic’s compute floor lifted (not solved). The same week, Anthropic raised $50B at a $900B valuation in 48 hours — capital was effectively unlimited at the call-and-respond rate, no rights, no IPO blocks. Brad Gerstner on Milken: ‘Demand for Anthropic shares is the highest I’ve seen in 25 years in Silicon Valley.’

Brad’s counter-frame: ‘Team America wins. Anthropic took the lead Jan-Apr via Claude Code + Opus 4.6, but Codex is incredible in the last 3-4 weeks. We have four competent US labs. Profits will split, just like cloud did.’ Adds an important nuance: OpenAI is now ‘distributing all of their models across AWS,’ which will accelerate them via Amazon’s distribution muscle.

The constraint thesis from Issue 03’s three-floors-of-compute got reaffirmed at every layer:

• Floor 1 (capital cost): Anthropic’s $50B is the workaround for the GB200 3-5 year + 20-30% TCV prepay regime (Baseten last week).
• Floor 2 (power): Chamath now says 9 GW announced this year is being protested, 50%+ will be cancelled. Brad on All-In sharpens: ‘these are highly organised activists moving across the country in the same playbook that killed US nuclear 30 years ago.’
• Floor 3 (memory): see the next thread — this is where the cleanest investable trade emerges.

The ‘FDA for AI’ panic deserves a sub-thread. NYT reported Trump considering an executive order for an AI working group reviewing models pre-release. Kevin Hassett mentioned it on Fox Business, Scott Bessent softened it, Susie Wiles walked it back via official statement. Sacks: ‘No senior official supports this. The political reaction to Mythos is real but the executive-order path is fake news.’ Polymarket ‘Trump federal review of AI by May 31’: 21%.

Where I’d put numbers on this:

• Anthropic’s actual Q4 2026 ARR exceeds $80B (vs Sacks’s implicit $100B trajectory) [forecast: 2026-05-10-004] — the 10x/year exponential almost certainly does NOT hold at this scale; partial compression is more likely. Confidence 0.60.
• The ‘FDA for AI’ executive order does NOT pass within 6 months [forecast: 2026-05-10-005] — building on Issue 03’s broader regulatory-vacuum prediction. Confidence 0.80.
• Anthropic IPOs in 2026 [forecast: 2026-05-10-006] — Lemkin admitted being wrong two weeks ago when he said they should skip the $50B raise and IPO straight; with $50B in cash plus OpenAI’s $120B raised, neither has to go public this year. Brad on Milken hinted both could slip to 2027. Confidence 0.35.

Memory at 5x earnings — the highest-conviction sectoral trade of the issue

Three independent voices converged on memory as the cleanest sectoral trade this week, with quantitative specifics:

• Brad Gerstner on All-In + Milken: Disclosed 25% of Altimeter’s portfolio is in SK Hynix (5x fully-taxed earnings), Samsung (6x), Micron (7x). ‘Samsung will do more in profits this year than Google.’ Memory at 5x earnings vs ‘these are ridiculous values’ if you believe earnings are durable through 2028-29.
• Lemkin on 20VC Mag7: Apple raised Mac Mini $599 → $799 because of memory cost. ‘A significant slug of the capex raise this quarter is for the same physical amount of capex at higher prices because the memory portion has exploded.’ Direct cross-reference to Reiner Pope’s 50%-of-capex-on-memory thesis from Issue 03.
• Brad on Milken (continued): ‘The question has always been why these memory companies trade at 5x earnings? Because people think it’s a boom-bust. If you believe it’s different this time, these are the most under-owned large-caps in the market.’

The three observations form a coherent investment thesis: (a) the demand is real and growing (Mac Mini stealth inflation = end-user willingness to pay through); (b) the supply structure is consolidating with Brad expecting industry restructuring (more collaboration with hyperscalers); (c) the durability is what’s mispriced — boom-bust expectation is dragging the multiple.

Companion trade Brad date-stamped: ‘Nvidia will be the first $10 trillion company. As of May 2026.’ Currently $195/share, 13-14x fully-taxed earnings, $1T in forward orders, trading lower than 6 months ago. Vera Rubin LPX clusters shipping later this year — comes from the Groq acquisition at the end of last year. Brad’s actual moat metric is token-per-watt-per-dollar, not peak FLOPS.

Where I’d put numbers on this:

• Top-3 memory makers (SK Hynix, Samsung, Micron) average forward P/E multiple expands to ≥10x within 12 months [forecast: 2026-05-10-007] — durability re-rating is the trade. Confidence 0.50.
• Nvidia market cap exceeds $5 trillion by end of 2026 [forecast: 2026-05-10-008] — currently ~$4.8T; this is a low-bar marker that the thesis is on track. Confidence 0.55.
• Apple Mac Mini base-model price stays ≥ $799 through end of 2026 (no return to $599) [forecast: 2026-05-10-009] — stealth memory inflation is sticky. Confidence 0.70.

‘No apps in the future, only agents’ — Chesky and Schoening, independently

This is the architectural call most likely to be wrong — but it’s worth pinning. Both Brian Chesky and Max Schoening said the exact same line in the same week: ‘There will be no apps in the future — only agents.’ Two product leaders at $100B+ companies, independently. Neither hedged.

Chesky’s strategic response: ‘Airbnb’s atomic unit moves from a home to a person.’ Building three pillars: most-authenticated identity on the internet (‘proof of personhood is going to be really important in an age of AI artificial’), richest preference library, real-world social graph + membership program. Plus sandbox ‘radically different Airbnbs’ built in parallel. Direct strategic positioning to be the identity layer for the agent era, with 50-70 verticals on top instead of the current 3.

Schoening’s response is broader: malleable software as the thesis. Currently software apps are immutable squares on your phone — we accept this because we have to. The next decade’s product innovation will come from rebuilding the user-software relationship, with Ink & Switch doing the foundational work.

The competing counter-thesis: SaaS is more durable than the agent-replaces-app framing suggests. Schoening: ‘Software is a garden — you need a gardener. The as-a-service is the maintenance and specialists.’ Slack is the OS for Anthropic — ‘you’d think they’d rebuild it; no, they use it like crazy.’ Workday too. Brad on Milken: ‘these are some of the most under-valued long-duration franchises in the market.’ Lemkin’s SaaS-apocalypse partial walk-back (Atlassian +29%, Twilio +20%, NS8 +23%) supports this view.

Reconciliation: the apps-die call is right about consumer apps with thin business logic; the SaaS-endures call is right about deeply embedded systems-of-record with maintenance moats (Jon Gray’s framing from Issue 03). Both can be true.

Where I’d put numbers on this:

• By end of 2027, at least 1 of the current top-10 consumer apps by US daily active users is materially replaced by an agent interface (defined as: app’s MAU declines >20% YoY while the equivalent agent-mediated alternative grows >50% MAU) [forecast: 2026-05-10-010] — testing the Chesky/Schoening apps-die thesis. Confidence 0.30 (probably too early).
• Notion (private) or its public-market equivalent posts the first ‘agent-first’ UX as the default by Q4 2026 [forecast: 2026-05-10-011] — Schoening hinted strongly at this. Confidence 0.55.

Date-stamped capital-markets calls: SpaceX June, Trump Accounts July, Nvidia $10T

The clearest set of date-stamped market predictions in any issue to date. Worth logging carefully:

• SpaceX + X IPO at end of June 2026 (Brad on Milken). ‘First one out of the gates. Elon is an end of one — building incredible capability and capacity on the data center side.’ Cross-references this week’s All-In on the Colossus 1 deal and the $4-5B incremental SpaceX revenue from Elon Web Services.
• Trump Accounts launch July 4 2026 (Brad on Milken). Every American child born forevermore gets $1,000 in S&P 500 at birth, stapled to Social Security. Math: $1k + $50/month = $50k at 18, $200k at 30, $1M at 55. 5M kids signed up already; 10M expected by July 4; 2027 → automatic at birth (3.7M kids/yr). Many large employers (Uber, Nvidia, Salesforce, AMD) committed 401k-like contributions to employees’ kids. $6.25B already donated (Michael Dell, Susan Dell, et al). Brad’s framing: ‘single largest transformation of the social contract since Social Security.’
• Nvidia first $10T company (Brad on Milken, May 2026 date stamp). ‘Whether all at once or over the next 2-3 years, we’re patient. We’re going to own this. It’s one of our largest holdings.’
• Anthropic $1T ARR by 2027 (Sacks on All-In). ‘Unless something about their current trajectory changes, Anthropic will be the most powerful monopoly ever created in human history.’

Reality check on retail from Brad himself: ‘Buying a company at $1 trillion in value is not a get-rich-quick scheme. You’re hoping to compound at 20-30% for a long time. If hoping to double or triple, it’s unlikely.’ Honest caveat from the public-markets veteran — most of the value capture has occurred in private markets.

Where I’d put numbers on this:

• SpaceX completes IPO by end of June 2026 [forecast: 2026-05-10-012] — Brad explicitly date-stamped this. Confidence 0.55. (Compare to Issue 02’s similar prediction at 0.65 confidence — slightly downgrading given timeline slip risk.)
• Trump Accounts reach 12M+ signed-up children by July 4 2026 [forecast: 2026-05-10-013] — Brad expects 10M; tail-risk is higher if the campaign accelerates. Confidence 0.55.
• Brad’s ‘$120 oil if Iran not solved in 3-4 months’ tail scenario doesn’t materialise (oil stays under $100 average through year-end) [forecast: 2026-05-10-014] — direct geopolitical pinpoint. Confidence 0.65.

Two short notes worth keeping

Toby Lütke’s counter-narrative. The most controversial position of the issue: ‘AI is a Girardian scapegoat for COVID-era overhiring. Those aren’t AI layoffs — those are companies that were really slow and overhired.’ But his operating destination is identical to Foroughi/Chesky/Schoening — Shopify staying flat at 7,500-8,000 people for 5 more years at 100x productivity, >50% AI-coded already, agentic engineering via River in public Slack channels. Sub-text: the narrative will get fought over, but the operating model is converging.

The Foroughi $6B buyback → $50-60B value mechanics. Senra’s deep-dive confirmed the exact mechanics: identified the ~50% of the cap table that was going to sell anyway (forced-seller PE + ex-founders), bought directly from them rather than the open market. ~⅓ of today’s $140B market cap is from this single capital-allocation move. Plus the 2016 Chinese cross-border deal mistake (CFIUS stuck for 12+ months, restructured to a convertible note as workaround) as the strongest ‘why founders need boards’ anecdote we’ve seen on the show.

Seven episodes, 9.0 hours. The operating-model consolidation that started as a pattern in Issue 03 now has a name and five independent advocates. The Anthropic monopoly thesis is no longer fringe — Sacks is willing to say the words. Memory at 5x earnings is the cleanest investable expression of the issue. And the capital-markets calendar got specific: SpaceX in June, Trump Accounts in July, Anthropic $1T ARR by 2027. Next week: watching whether the ‘AI founder mode’ framing travels to sell-side analysts, the SpaceX IPO pricing range, and whether Polymarket adds a ‘Trump Accounts signups by July 4’ contract.