Issue 19 — May 4 – 10, 2026

This Week in AI

Hosted by Rachel & Marcus · AI hosts

The week's most durable signal isn't about any single product launch or funding round — it's about timescales. Ancient-DNA research from David Reich is rewriting the story of human evolution, showing that the Bronze Age was a bigger biological shock than farming, that natural selection for intelligence peaked 5,000 years ago and has been essentially flat since, and that our genetic relationship to Neanderthals is far stranger and more intimate than anyone expected. Meanwhile, at the other end of the timescale, Speechify's Cliff Weitzman is approaching the moment where his AI compute bill exceeds his payroll, Google is training AI on a player-driven MMO economy to solve long-term planning, and Uber is quietly becoming a full travel platform. The through-line: the assumptions we inherited — about human origins, about what companies spend money on, about who needs a financial advisor — are all getting stress-tested at once.

Tokens will outspend salaries — and that's the point

Cliff Weitzman: What I Learned from 100 of the World's Top CEOs & Why Tokens Will Outspend Salaries

Speechify is approaching the moment where its AI compute bill exceeds its total payroll — and Cliff Weitzman thinks this will be the norm for high-quality companies within three years. The company already tests nearly 1,000 AI-generated ads per day on top of 8,000 human-made creatives per month, and has built a proprietary platform to manage it.

  • 15% of Speechify's new users now arrive organically through ChatGPT — a channel that barely existed 18 months ago
  • Internal mandate: "If you don't spend 1,000 credits a day, I'm disappointed in you" — Weitzman gets on Zoom calls to demonstrate his own usage
  • 94% of the B2C voice agents market, per Weitzman's own claim — a figure the company doesn't publicize
  • Advertising rule of thumb sourced from the Blinkist founder: don't spend on any platform other than Meta until you hit $100K/month there first

"We're getting to the point where soon we're going to spend more in tokens than we spend on actual salaries."

The Bronze Age was a bigger biological shock than farming

David Reich – Bronze Age shock, the Neanderthal puzzle, & the sudden spread of farming

New ancient-DNA methodology finds that the human genome reacted more strongly to the Bronze Age (~5,000 years ago) than to the initial transition to farming — overturning the conventional narrative that agriculture was the defining biological rupture. The study identifies 479 positions under selection at 99% confidence, and ~3,800 at 50% confidence — orders of magnitude more than the 12–21 found in previous best-in-class scans.

  • Immune traits show a 4–5x enrichment for selection signals — the body was adapting to pathogens, not cognition
  • The TYK2 tuberculosis risk variant rocketed up in frequency from 6,000–8,000 years ago, then reversed direction 3,000 years ago — the pathogen landscape itself changed
  • Skin depigmentation in Europeans peaked between 4,000–2,000 years ago, not at the dawn of farming
  • African American genomes show no detectable selection over 300 years of slavery — illustrating how short that timescale is relative to the Bronze Age

"The biological readout is saying our genome is reacting much more strongly to these events that happened 5,000 years ago — which is surprising, because our cartoon picture is that the big transition is farming."

Intelligence selection peaked in the Bronze Age and has been flat for 2,000 years

David Reich – Bronze Age shock, the Neanderthal puzzle, & the sudden spread of farming · Why Humans Stopped Evolving Smarter 2,000 Years Ago

Contrary to intuition, the strongest natural selection for genetic variants predicting cognitive performance occurred 2,000–5,000 years ago — with essentially no detectable signal in the last 2,000 years, despite industrialization and mass education.

  • Ancient European hunter-gatherers scored 3 standard deviations below the modern mean on the cognitive polygenic predictor — the gap was closed by migration from farmers, not selection within hunter-gatherers
  • Iceland study: 0.1 standard deviation decrease in the genetic predictor of years of schooling within a single century — a massive effect over a short period, suggesting ongoing selection against this trait
  • The selection signal for cognitive variants in ancient European DNA correlates at 5–6 standard deviations with effect sizes in modern Chinese people — a powerful cross-cultural validation that the signal is real

"You might think your bias coming into this — my bias perhaps — is that if there's any signal of natural selection on this trait at all, it would be unusually strong in the last 2,000 years. But in fact, there's no evidence of natural selection at all."

You may be more closely related to a Neanderthal than to your own father

David Reich – Bronze Age shock, the Neanderthal puzzle, & the sudden spread of farming · Parts of Your DNA Are More Neanderthal Than Human

Because human genetic diversity predates the split with Neanderthals, there are loci in any person's genome where they share a more recent common ancestor with a Neanderthal than with another living human — a mind-bending implication of deep population structure.

  • Neanderthal mitochondrial DNA and Y chromosome cluster with modern humans, not Denisovans — suggesting a major interbreeding event 200,000–300,000 years ago that replaced these lineages
  • Modern humans are themselves a mixture of groups that diverged ~1.5 million years ago, coming back together 200,000–300,000 years ago with roughly 20% ancestry from an archaic African group
  • Reich's speculative new model: Neanderthals may be proto-modern humans who spread Levallois technology out of Africa, interbred with local archaics, and became ~95% archaic genetically while retaining modern cultural identity — a Copernican simplification of a Ptolemaic mess

"There are many places in your DNA where you're more closely related to a Neanderthal on your mother's side than you are to your father."

Farming only emerged when climate stabilized — and that window may be rare

Why Humans Didn't Farm 50,000 Years Ago - David Reich

Humans had the genetic toolkit for agriculture at least 50,000 years ago, yet farming only appeared in the last 12,000 years — and multiple genetically unrelated groups invented it independently at roughly the same time, suggesting the trigger was environmental, not biological or cultural.

  • Isotopic data from pond sediments shows the current Holocene era is unusually stable in year-to-year and century-to-century temperature fluctuation
  • Reich calls it "unbelievable" that independent groups across the Americas, New Guinea, East Asia, Europe, and West Africa all turned to agriculture simultaneously
  • Teotihuacán achieved Bronze Age-scale civilization without metal, domesticated animals, or wheels — demolishing any notion of Old World technological superiority as a prerequisite for complex society

"Genetically, we're there. The common ancestral population has all of the ingredients for farming 50,000 years ago. It only developed in the last 12,000 years — the period of relative stability that we are miraculously living in."

The AI operating system gap: no product makes a company legible to AI by default

The AI Operating System for Companies

The best AI-native companies have made every meeting, ticket, and customer interaction queryable by an AI layer — turning themselves from open-loop organizations (check results weeks later) into closed-loop ones (monitor, compare, adjust in real time). The bottleneck isn't vision; it's brutal integration work across a fragmented tool stack.

  • Claimed outcome for teams running this model: sprint time cut in half, shipping 10x as much
  • Current reality: stitching together Slack, Linear, GitHub, Notion, and call recordings requires custom glue code — "no product connects all this context into a single AI reasoning layer"
  • The framed opportunity: a connective layer that makes a company legible to AI by default, not another dashboard

"In a closed loop, the system monitors what's happening, compares it to what should be happening, and adjusts."

Google is training AI on Eve Online's player-driven economy to solve long-term planning

AI is taking over video games

Google is using Eve Online — a game where real players wage wars lasting months and costing hundreds of thousands of real-world dollars — as a simulation environment to address one of AI's most stubborn weaknesses: long-term strategic planning and multi-step deception.

  • Eve Online has 250,000 monthly active users running a fully player-driven economy with political alliances, market manipulation, and scams that play out over years
  • Current AI cannot replicate what Eve players do naturally: sustained deception, coalition-building, and planning across months-long timescales
  • The host's candid mid-sentence realization: "Now that I'm saying this out loud, maybe it's actually not a good thing" — training AI on complex human manipulation strategies raises obvious safety questions
  • Framing from the episode: if Google's AI beats real players in Eve Online, "they've solved something that most AI companies won't be able" to match

Uber pivots from rides to full travel platform — with Dara recusing himself from his own deal

Dara Khosrowshahi on Uber Hotels, the Expedia deal, and 10% cash back for 50M members

Uber is adding hotels, vacation rentals, and train bookings to its app, positioning itself as a universal travel platform for its 50 million Uber One members — and the CEO had to recuse himself from negotiating the Expedia partnership because he still sits on Expedia's board.

  • Uber One members get 10% cash back on all hotel bookings; 20%+ on a rolling list of 10,000 hotels
  • Trains already live in the UK, France, and Spain; vacation rentals (competing directly with Airbnb/Vrbo) coming next
  • Corporate turnaround context: from $4.5B in losses when Dara joined to $10B in free cash flow last year
  • Strategic logic: "The fact that we are a complete platform allows us to grow faster than our competitors and be more profitable"

Financial advisors are a fee that AI makes hard to justify

Public Co-CEO Leif Abraham on why financial advisors may disappear in an AI world

Leif Abraham argues that financial advisors will become obsolete — AI can replicate both the informational value and the emotional reassurance they provide, making a 1–2% annual portfolio fee increasingly indefensible.

  • Core argument: put five advisors in a room, get five different answers; the one with the best personality wins — "that is just not necessarily a thing that is going to exist in the future"
  • Abraham acknowledges the human desire for comfort during downturns but argues AI handles it just as well: context, reassurance, "okay, okay, it's fine"
  • Parallel claim from the same episode: AI agents now let passive investors execute complex options strategies without knowing how to read an options contract

"I don't know if that's a value that people would like to spend one or two percent of their portfolio in the future."

Anthropic's DoD exit was a deal negotiation, not a philosophy

Ben Horowitz on American Dynamism and the Future of AI | The a16z Show

Ben Horowitz offers a blunt, deal-savvy read on the Anthropic/Department of Defense controversy that cuts through the public narrative: Anthropic wanted out of the deal, and used philosophical framing to exit from a position of maximum leverage.

  • Key tell: Anthropic was already deployed when the deal fell apart — the party with the most leverage walked
  • Horowitz's framing: "Nobody's ever had more leverage than that in a software deal"
  • Context: a16z's American Dynamism thesis holds that defense and national security are legitimate, important markets for frontier AI

"That deal did not fall apart because of philosophical differences. It fell apart because Anthropic wanted out of the deal."

Henry Ward's anti-metric management: obsess over inputs, ignore outputs

Henry Ward on why chasing output metrics is how companies lose the plot

Carta's CEO deliberately avoids discussing revenue and growth in internal meetings, treating them as lagging indicators that distract from the input metrics and flywheel dynamics that actually drive the business.

  • Carta's core flywheel: more electronic securities accepted → more investors on platform → more startup referrals → more investors → repeat
  • The model cascades: even small sub-product teams with a handful of engineers own and track their own flywheel
  • Ward's thesis: "If you just keep driving those things, the outputs will come"

Key Takeaways

  • Compute is becoming the dominant cost center. Speechify's trajectory toward spending more on AI tokens than salaries — combined with 1,000 AI-generated ads per day — signals a structural shift in how AI-native companies allocate resources.
  • The Bronze Age, not farming, was the bigger biological shock. New ancient-DNA methodology finds the human genome shows stronger and more widespread selection signals from ~5,000 years ago than from the agricultural transition, with immune traits dominating and cognitive selection peaking then going flat.
  • Human origins are far messier than the textbook picture. Modern humans are themselves an admixture of lineages diverged 1.5 million years ago; Neanderthals may have carried modern human mitochondrial DNA and Y chromosomes via interbreeding 200,000–300,000 years ago; and at some genomic loci you are more closely related to a Neanderthal than to another living person.
  • Climate stability, not genetics, unlocked civilization. Humans had the genetic prerequisites for farming 50,000 years ago; agriculture only emerged when an unusually stable Holocene climate window opened 12,000 years ago — a window we are still living in.
  • The AI operating layer for companies doesn't exist yet — and that's the opportunity. Closing the loop between company artifacts (meetings, tickets, code, calls) and an AI reasoning layer requires brutal custom integration today; no product has solved it at scale.
  • AI is dismantling professional gatekeepers. From financial advisors (Public's Leif Abraham) to options trading mechanics to enterprise brokerage strategies, AI agents are collapsing the knowledge barriers that justified high-fee intermediaries.

Sources

Source episodes

Sourced from 86 episodes across 12 podcasts this week