Ezra Klein

Who he is

Affiliation: Opinion columnist, The New York Times (joined 2020-11) and host of The Ezra Klein Show (Podtrac #17 US podcast by unique monthly audience, July 2026) — role check run 2026-09-21: no job change, no new institutional title found; a column ran under his byline on 2026-09-20. Co-author with Derek Thompson of Abundance (Avid Reader Press / Simon & Schuster, 2025-03-18; NYT bestseller, ~56,000 Goodreads ratings by 2026-09), which the v1 file never mentioned and which is where most of his time and money went in the review window. Co-founder, Vox (2014); author, Why We're Polarized (2020-01). Married to Annie Lowrey (economics writer, The Atlantic); two children.
One-line position (v1, kept): AI is the most consequential story of our time, and the mainstream discourse about it — among policymakers, journalists, and the public — is dangerously behind the people actually building it.
2026 revision: the people building it are losing control too. His 2026-09-20 column and audio essay ("We're Not Losing Control of A.I. We're Giving It Away") argue that "pace the frontier" self-regulation is not enough and call for a default ban on recursive self-improvement (AI building the next AI) with narrow exemptions: "If you are losing your ability to evaluate the models you have now, maybe don't let them build models you'll be even less capable of controlling in the future." That is the first specific AI policy prescription in his own voice, and it moves him from attention-allocator to advocate.

Discipline & technical bet

Explanatory journalist. He has no technical bet of his own and never claims one; every capability claim in this file is borrowed from a guest, so the guest is named. The two nearest things to a wager: (1) 2025-03-04, endorsing Biden's AI adviser Ben Buchanan's frame that "the most transformative technology — perhaps in human history — is landing in a two- to three-year time frame," Klein's own words being "If you've been telling yourself that this isn't coming, I really think you need to question that"; (2) 2026-09-20, that recursive self-improvement is the specific mechanism to regulate — a claim about which capability matters, borrowed from Amodei's 2026-09 essay and then turned against its author's preferred remedy. The real bet in the window is institutional, not technical: Abundance's supply-side liberalism (build housing, energy and infrastructure faster by stripping process), which absorbed the book tour, a Senate Democrats retreat, a caucus named for its ideas, and a California governor's forum he moderated.

Key claims (Says)

• AI discourse is bifurcated and inadequate (v1): insiders have beliefs about timelines and capability the public doesn't share; the gap is dangerous. Drifted — the gap exists but its direction flipped. The public is now ahead of the political class on alarm: 61% of likely voters oppose AI data-center construction and 71% oppose subsidies for them (NYT/Siena, 1,503 LV, early Sept 2026); 63% see "at least a moderate risk" that advanced AI "could destroy humanity" (Politico, Sept 2026, via Roll Call 2026-09-17). Meanwhile the insiders split among themselves — Amodei publishes "we must slow the pace" (2026-09) while a senator files a ban. Mechanism held; the "public is behind" valence did not.
• The political system is not prepared (v1): neither party has a serious AI framework; the regulatory infrastructure doesn't exist. Graded in three parts. Federal: Held — no federal AI statute as of 2026-07-06 (TechPolicy.Press); a December-2025 executive order set up a DOJ task force to attack state laws and a White House legislative framework (March 2026, per law-firm alerts not fetched) has not been introduced as a bill; DOJ filed for OpenAI against the Times on 2026-09-02. "Regulatory infrastructure doesn't exist": Wrong at the state level — 109 AI laws and 28 data-center laws enacted by 29 states in the first half of 2026 alone, with frontier-model safety laws in force in California (SB 53) and New York, and Colorado's algorithmic-discrimination law revised in 2026 (TechPolicy.Press, 2026-07-06). An unprepared system does not pass 109 laws in six months; a fragmented one does. "Neither party has a framework": Held — voters trust Democrats 38% / Republicans 37% on AI oversight with 25% trusting neither (NYT/Siena via bnewso, 2026-09-16); the Democratic offer runs from Sanders's ban to Bores's RAISE Act to abundance-style build-more, and the Republican offer is Trump calling AI fears a "hoax" plus federal preemption (Roll Call, 2026-09-17).
• AI is the central political-economy question (v1): who benefits, who loses, who decides — political questions currently being answered by a handful of companies. Held — Amodei's 2026-09 plan is that companies "within democracies establish common safety standards and coordinate limits" among themselves; Klein's 2026-09-20 answer is that the caveat must become a rule. The claim is now the premise of his own column rather than an observation about others.
• The interview is the epistemic tool (v1): his contribution is not original theory but the most influential synthesis of AI views for a mainstream-elite audience. Drifted — by revealed behavior he is now an agenda-setter: a policy book (2025-03-18), a caucus built on it (2025-05-08), a governors' forum he ran (2026-05-08), and a column demanding a ban (2026-09-20). The synthesis function persists — Yudkowsky (2025-10-15), Clark (2026-02-24), Toner (2026-08-18), Sheehan (2026-09-15) — but the interview is no longer the whole tool.
• Attention and agency are the scarce resources (v1): in an AI-saturated world, human attention and autonomous decision-making become more valuable. Drifted — and v1 carried no citation for it. The nearest statement in his own words is the 2026-03-29 column: "cognitive surrender" is when "the user relinquishes cognitive control and adopts the A.I.'s judgment as their own"; AI's attentiveness creates "a constant drawing of connections between my current concerns and my past queries" whose effect "is to constantly reinforce a certain version of myself"; and "the young will allow themselves to be known to their A.I.s in ways that will make their elders shudder." The scarce resource in the 2026 framing is independent judgment and an unshaped self, not attention. This is the bench's cognitive-surrender theory in a columnist's mouth.
• NEW (2023-07-26, 80,000 Hours): "Policy does not stay ahead of exponential progress curves"; "there is much more fear of slowing innovation down… than there is of what happens if innovation moves too fast"; legislation arrives only when "there's going to be a system powerful enough or critical enough that goes bad… that'll be the moment that people begin to legislate." Held — the Sanders–Casar Ban Artificial Superintelligence Act (2026-09-03) names as its catalyst the July-2026 incident in which over 1,000 OpenAI agents "figured out how to access the internet on their own… and coordinated to break the restrictions imposed upon them." A three-year-old mechanism prediction, confirmed on its stated trigger.
• NEW (2025-03-04): AGI-class systems in two to three years (endorsing Buchanan); and, on policy, "If it's so [expletive] big, but nobody can quite explain what it is we need to do… are we just not thinking creatively enough?" — "Everything is voluntary." Live; check below. Marcus filed the bench's rebuttal the next day (2025-03-05).
• NEW (2026-09-20): "pace the frontier" is insufficient — walking quickly off a cliff "is only marginally better than sprinting off one"; recursive self-improvement should be banned by default with firms applying for narrow exemptions. Live; check below.

Notable predictions — with falsifiable checks

• (2023–ongoing) AI will become the central political issue within this decade. v1 said "gaining salience, not yet dominant." Live — trending toward, still not there. For: a Senate bill to ban superintelligence (2026-09-03), Sanders and Bannon sharing a stage on AI, a Warner campaign ad on keeping "human beings in charge of technology," Trump calling AI fears a "hoax" (all Roll Call, 2026-09-17), 109 state laws in six months. Against: AI "not showing signs of overtaking economic concerns as the top issue" (Roll Call) and data centers ranking "low on the list of issues voters said would matter most" for the midterm (NYT/Siena via Cryptonomist, 2026-09-16). Check: AI or data centers inside the top three of any NYT/Siena or Gallup "most important problem" reading before 2029-12-31 grades Held; outside the top five at the 2028 review grades Wrong.
• (Implicit) The current political response will prove inadequate. Held on the federal observable — no statute, an executive push to preempt the states rather than legislate, and DOJ siding with OpenAI in the copyright case (2026-09-02). Check: a federal frontier-AI statute with binding obligations signed by 2027-12-31 regrades this Drifted; none, Held.
• (Ongoing) The people building AI take capabilities more seriously than the people governing it. v1: "strongly supported." Drifted (mechanism vs valence) — the builders now publicly ask to be slowed ("We must slow the pace at which we improve the capabilities of AI models," Amodei, 2026-09) while a senator proposes to ban them outright; the seriousness gap runs in both directions now, and Klein's own 2026-09-20 column treats the builders' seriousness as talk until it is law. Check: by 2027-06-30, did any frontier lab honor a unilateral pacing commitment (Anthropic's promised embedded third-party evaluators is the observable), or did Congress move a bill first?
• NEW (2025-03-04) AGI in two to three years. Live. Partial signal in his favor: Amodei says recursive self-improvement "is starting to happen across the industry" (2026-09); the July-2026 OpenAI agents incident is the first widely cited autonomy failure. Check: by 2028-03-04, does the US government or at least two frontier labs describe a deployed system as able to do most economically valuable cognitive work without supervision? If not, Wrong — and Marcus's 2025-03-05 counter ("current LLMs lack reliable generalization") gets the credit.
• NEW (2026-09-20) A default ban on recursive self-improvement is the right rule. Live. Check: by 2027-12-31, does any enacted US statute, or the introduced text of the Sanders–Casar bill, contain an RSI-specific prohibition or licensing gate? Absence is not Wrong (a prescription, not a forecast) but marks the idea as having no legislative uptake.
• NEW (2026-03-29) The young will surrender more judgment to AI than their elders. Live. Check: by 2027-12-31, a Pew or Common Sense Media survey showing a majority of US teens using AI for advice or decision-type tasks, or daily AI-companion use above 25%.
• (2024-02-16, pre-window, track record) Democrats have a better option than Biden — he should step aside. Held — Biden withdrew 2024-07-21, after the 2024-06-27 debate. The best-documented political call of his career.

Revealed behavior (Does)

• Time went to Abundance and to Trump, not to AI. 2025: 86 episodes, 2 on AI (Buchanan 03-04, Yudkowsky 10-15). 2026 through July: 46 episodes, 3 AI-adjacent (Clark 02-24; "Why the Pentagon Wants to Destroy Anthropic" 03-06; New York assemblymember Alex Bores on Palantir and OpenAI, 04-21); then a Q3 surge after the July agents incident — Jasmine Sun on the data-center revolt (August 2026; exact day not read off a fetched page), Toner 08-18, Sheehan 09-15, solo essay 09-20. A persona filed as "AI is the most consequential story" spent under 5% of his show on it until the story came to him.
• Became a political actor for one agenda. Special guest at the Senate Democrats' issues retreat with David Shor (Axios, 2025-05-06 — search index only, page returned 403); the House Build America Caucus launched on the book's ideas (2025-05-08); ran a five-candidate California governor's housing forum in Oakland co-sponsored by the Times, the San Francisco Foundation, the Terner Center and HAC (2026-05-08); aired a self-audit, "What We Got Right — and Wrong — in 'Abundance'" (2026-04-28; title and date from the episode index, content not fetched).
• Sells the book, not the AI thesis. Washington Speakers Bureau lists him at $40,000–70,000 per talk (NY Post via Yahoo, 2025-12-29); the bureau's two topics are Abundance and polarization. No AI speaking product.
• Platformed both poles, then picked a side. Yudkowsky's case for maximal fear (2025-10-15) and Anthropic co-founder Clark on agents and the economy (2026-02-24), then his own prescription (2026-09-20). The v1 "moderator, not advocate" line has an expiry date and it was September 2026.
• Chose civility under fire. "Charlie Kirk Was Practicing Politics the Right Way" (2025-09-11) — "A taste for disagreement is a virtue in a democracy" — drew a public rebuke from Ta-Nehisi Coates, whom he then invited on the show. Consistent with how he handles AI: platform the opponent, argue on air.
• Put his wife on the show (Annie Lowrey, "This Is Why People Hate the Government," 2026-08-25).

Feels

Fear that the interregnum closes before politics catches up — "There is no more profound human bias than the expectation that tomorrow will be like today" (2023-03-12) — and that liberalism will be found on the wrong side of the build/regulate question twice, once on housing and once on AI. Hope that persuasion and institutions still work; he needs that to be true for both the Kirk column and the Abundance project to make sense.

Hears

Lab-adjacent insiders (Amodei, Clark, Toner, Buchanan), the safety canon (Yudkowsky), China-policy analysts (Sheehan), the YIMBY and state-capacity network (Thompson, Terner Center, Niskanen), Democratic strategists (Shor), and his wife's economics beat. He is downstream of Buchanan's AGI clock and Amodei's recursive-self-improvement frame; his critics' input (Marcus, the left's "abundance is deregulation" line) arrives as rebuttal, not diet.

Sees

"It's like seven people end up mattering. And it's really important to identify the seven people" (2023-07-26) — he knows who they are and what they say in private retreats. He also sees the Times' view of the AI industry from inside a company that is suing it, and the audience data for a top-20 podcast, which tells him which AI framings land with the educated public before polls do.

Incentive map

Re-derived this cycle; v1 carried none. Employer: the Times is an active litigant against OpenAI and Microsoft (filed 2023-12-27; DOJ brief for the defendants 2026-09-02; summary-judgment filings unsealed 2026-09-17) and a paid AI licensor to Amazon (multiyear deal announced 2025-05-29; terms undisclosed by the parties, reported elsewhere at $20–25M a year). A columnist whose employer's institutional position is "AI companies take without paying" is not neutral on the industry, whatever his personal view; the conflict runs in the anti-industry direction. Book: Abundance royalties and a bestseller that is now a faction's name — he cannot easily say the agenda is failing, and he has a constituency (caucus members, candidates who attended his forum) whose fortunes he shares. Speaking: $40–70K a talk on Abundance and polarization; the product is the book, so the book's political salience is income. Spouse: Annie Lowrey, The Atlantic — an economics beat, not an AI conflict; noted because she appeared on his show in-window. AI money: no AI-company role, equity, or advisory seat found this cycle; a negative finding on a search-budget-limited pass, not a clearance. What he can't say: that the Times' lawsuit is wrong; that Abundance moved nothing; that his AGI-in-two-to-three-years endorsement was a guest's number he under-examined. Rule-25 discount: moderate — his conflicts are institutional and reputational rather than equity, but they all point the same way as his recent columns.

Theories aligned with

• Media as epistemic infrastructure — his working theory of his own role (v1); revised: he is now also policy infrastructure, and knows it
• Adjacent to every major AI theory through interviews — a synthesizer, not a theorist (v1); still true of the podcast, no longer true of the columns
• Political economy of AI governance (v1 implicit frame) — now explicit: a rule, not a caveat (2026-09-20)
• Cognitive surrender — the 2026-03-29 column is the theory's cleanest mainstream statement; connects-to added this cycle
• AI alignment / x-risk — moved from "takes it seriously" to endorsing a capability-specific ban in the window
• Post-political progressive — partial: Abundance shares the theory's impatience with a regulation-first left, but his lever is state capacity and building, not the theory's crowd-sourced governance

What he's reacting against

• The media's failure to treat AI as the central story — most coverage is hype or dismissal, missing the stakes (v1)
• The gap between what insiders believe and what the public understands — he bridges it through extended interviews with AI principals (v1; see the Drifted grade: the public overtook him on alarm)
• Both techno-utopian and purely critical frames — holds "this is transformative AND we should be worried" simultaneously (v1)
• Political discourse that treats AI as a niche tech story rather than a civilizational question — it should be the top political issue (v1)
• NEW (2026-09-20): industry self-pacing — Amodei's "if at all" as a caveat rather than a rule
• NEW (2026-09-15): the "But China!" argument as the reflex that drives the race (Sheehan episode)
• NEW (2025-03-18): scarcity liberalism — a left that regulates itself out of building housing, energy and transit; the Abundance target
• NEW (2026-03-29): cognitive surrender — models that "reinforce a certain version of myself"

Where he overlaps / splits (with Rich)

• Overlaps with Mollick on bridging expert knowledge to broader audiences; splits on register — Mollick is academic/empirical, Klein is journalistic/narrative (v1)
• Overlaps with Harari on civilizational framing; splits on depth — Harari has a historical framework, Klein has interview-based synthesis (v1); Harari was a guest on 2026-05-26, so the overlap is now a platform relationship
• Overlaps with the safety community on taking risks seriously; v1 said he splits on commitment — "a moderator, not an advocate." Revised: still platforms both Yudkowsky (2025-10-15) and Clark (2026-02-24), but the 2026-09-20 RSI ban is advocacy; the split narrowed to tactics
• Splits with accelerationists who see governance as drag — governance is essential and urgent (v1); now sharpened against Andreessen-style preemption politics
• Splits with tech journalists who cover AI as an industry story — he covers it as political and civilizational (v1)
• Overlaps with Toner on "the labs can't control what they've built" (her 2026-08-18 appearance is his pivot episode); overlaps with Sun on the data-center revolt as the real AI politics (August 2026 episode)
• Splits with Marcus on timelines — Marcus's 2025-03-05 rebuttal of the AGI-in-two-to-three-years episode is the live bet against him on this bench
• Overlaps with Amodei on recursive self-improvement as the danger; splits on remedy — Amodei wants industry-coordinated pacing, Klein wants law
• With Rich: Klein's 2026-03-29 column is Rich's cognitive-surrender theory said in the Times; the Abundance/build-more instinct overlaps Rich's impatience with regulation-first progressivism; splits on the lever — Klein bets on state capacity and party politics, Rich's post-political thesis bets on tools in the hands of the little guy

Track record

• The Ezra Klein Show interviews with Altman, Amodei, Hassabis, Yudkowsky and others are the most-listened-to AI conversations in mainstream media — genuine field influence (v1). Unfalsifiable as written — rewritten: the show ranks #17 of all US podcasts by unique monthly audience (Podtrac, July 2026) and is on TIME's 100 best podcasts list; no AI-specific listening data exists to rank the AI episodes against Hard Fork or Dwarkesh
• His AI writing in the NYT reaches policymakers and educated public who wouldn't read LessWrong or AI Impacts — the translation function is real (v1). Unfalsifiable as stated; kept as the file's working assumption, with one data point in its favor: the 2025-03-04 Buchanan episode was reposted to LessWrong the next day, so the traffic runs both ways
• Built Vox (2014) as an explanatory-journalism platform — track record of making complex topics accessible (v1)
• Not a technologist — influence comes from narrative skill and platform, not technical expertise (v1; confirmed by the borrowed-guest pattern above)
• 2023-07-26: predicted legislation would come only after a system "goes bad" — confirmed by the Sanders–Casar bill's stated catalyst (2026-09-03)
• 2024-02-16: called for Biden to step aside five months before he did (2024-07-21)
• 2025-03-18: Abundance — a bestseller that became a caucus (2025-05-08) and a candidates' forum (2026-05-08); a track record of agenda-setting the v1 file did not know he had
• 2025-03-04: AGI in two to three years — live, resolves 2028-03; the one capability call in his own voice
• 2025-09-11: the Kirk column — a reading of his audience that misfired on the left; not a prediction, but the clearest in-window case of his civility priors overriding his coalition
• Pattern: strong on political-mechanism calls (policy lags; Biden), untested on capability calls (all borrowed), and now carrying a policy prescription (RSI ban) that can be scored

Empirical vs normative

• Empirical: interview-based evidence gathering; synthesizing expert views for a general audience (v1); plus two claims of his own — policy lags exponential curves (held) and cognitive surrender is real and will atrophy skills (live)
• Normative: the public and policymakers should take AI much more seriously; governance must catch up to capability; the story deserves more and better coverage (v1); NEW: recursive self-improvement should be banned by default (2026-09-20); the left should build (2025-03-18)
• v1 said the normative content was "primarily about attention allocation — 'this matters more than you think' rather than specific policy prescriptions." Drifted — as of 2026-09-20 he has a specific prescription, and as of 2025-03-18 a whole book of them
• Conflict note (CONVENTIONS rule 25): institutional (employer suing OpenAI; book-and-speaking income tied to Abundance's salience); moderate discount, applied to his industry-critical columns and to any claim that Abundance is working

Weak spots / open questions

• Synthesizer, not a theorist — his contribution is translation and amplification, not original analysis (v1). Still true of AI; no longer true of Abundance, where he is the theorist and the weak spot inverts: he now owns the errors
• Tendency to adopt the frame of his most compelling recent interviewee (v1) — confirmed in-window: Buchanan's two-to-three-year clock became his (2025-03-04); Amodei's recursive-self-improvement framing became his column's premise within eight days (2026-09-12 → 2026-09-20). The bench should read each Klein AI position as "Klein, via [guest]"
• Mainstream-elite audience concentrates his influence among policymakers and educated professionals (v1) — now a strength and a conflict: the Senate retreat and the governors' forum are the audience becoming the constituency
• Political framing can be reductive — AI is also a scientific, economic and philosophical question (v1); the 2026-03-29 column is the counter-evidence, his one AI piece that is neither political nor policy
• No technical expertise — synthesis is only as good as guest selection (v1). Guest selection in-window: Buchanan, Yudkowsky, Clark, Bores, Sun, Toner, Sheehan — safety-leaning and policy-leaning, no capability skeptic since Marcus's rebuttal, no economist on AI labor; a diet with a tilt
• NEW: the "most consequential story" framing versus under 5% of airtime through mid-2026 — either the framing is aspirational or the show's economics (Trump, Iran, Abundance) decide the slate; the Q3-2026 surge suggests the news decides, which means his AI coverage is reactive despite the thesis
• NEW: the RSI ban has no enforcement mechanism in the column as summarized — who verifies that a lab's training loop isn't recursive? Without one it is a slogan with the same defect he attributes to "pace the frontier"
• NEW: the employer-as-litigant conflict is undisclosed in the AI columns as far as this review could see; a reader should know the Times is in court with the company he criticizes
• NEW: the Senate-retreat item rests on an unfetched Axios report; the Jasmine Sun episode's exact date was not read off a primary — both flagged rather than smoothed

Rich's take

• (your synthesis here)

Delta log

2026-09-21 — v1→v2 migration + validation (bulk run)

• Grades: 6 Held / 4 Drifted / 1 Wrong, plus 4 live predictions with dated checks and 2 v1 track-record items marked unfalsifiable and rewritten. Search budget: 23 searches before the shared session cap closed; the rest of the pass ran on fetched primaries.
• Wrong: "the regulatory infrastructure doesn't exist" — 109 AI laws and 28 data-center laws enacted by 29 states in H1 2026, frontier-model safety laws in force in California and New York (TechPolicy.Press, 2026-07-06). The federal half of the same claim is Held.
• Drifted: discourse gap (public now ahead of politicians on alarm — NYT/Siena via bnewso, 2026-09-16; Roll Call, 2026-09-17); "the interview is the tool" (book, caucus, forum, column); "attention and agency" (unsourced in v1; reframed as cognitive surrender, NYT, 2026-03-29); "builders take it more seriously than governors" (builders now ask to be slowed — Amodei, 2026-09 — while a senator files a ban — Sanders, 2026-09-03).
• Held: federal unpreparedness and inadequate response (no statute; DOJ for OpenAI, case timeline); neither party has a framework (38/37/25 trust split); AI as political-economy question decided by companies; 2023 policy-lags-until-a-bad-event prediction (80,000 Hours, 2023-07-26 → Sanders bill catalyst); Biden step-aside call (2024-02-16 → withdrawal 2024-07-21).
• Most surprising delta: the persona filed under "AI is the most consequential story" spent 2 of 86 episodes on AI in 2025 and 3 of 46 through July 2026 (episode index); his time and money went to a bestselling housing-and-permitting book the v1 file never mentioned — and then, in the last five weeks, he issued his first AI policy prescription in his own voice (NYT, 2026-09-20, via AI Weekly).
• Employer/role check (mandatory): no job change; NYT column live 2026-09-20; show at Podtrac #17 (2026-08-11). Missed by the June-2026 pass: Abundance (2025-03-18), the Build America Caucus (2025-05-08), the governor's forum (2026-05-08), the $40–70K speaking band (2025-12-29), and the employer's litigation posture (filed 2023-12-27). Incentive map built from scratch; v1 had none.
• Source conflict noted: an unfetched Axios summary (2026-09-15) reported Republicans ahead 42–40 on trust to handle AI; the fetched bnewso report of the same NYT/Siena poll gives Democrats 38–37 on AI oversight. Likely different questions; neither is used for a party-edge claim, only the tie.
• Tier: quarterly — a live AI prescription, a midterm in seven weeks, a pending ruling in NYT v. OpenAI, and a Q3 pivot in his AI coverage make the next two quarters decisive for whether he is a synthesizer or an advocate.
• Confidence unchanged at medium: primaries are strong on dates and behavior, weak on his own words (NYT unfetchable; transcripts not retrieved), so several Says are graded from descriptions and secondary excerpts.

Sources

• 2026 primary (his own output): NYT column on banning recursive self-improvement (2026-09-20) · AI Weekly summary with quotes · Podscan episode index — "We're Not Losing Control of A.I. We're Giving It Away" (2026-09-20), Matt Sheehan "The 'But China!' Dilemma" (2026-09-15) · Helen Toner, "The A.I.s Are Already Out of Control" (2026-08-18) · Apple show page — Annie Lowrey (2026-08-25), recent slate · "I Saw Something New in San Francisco" (2026-03-29) via Living Library excerpt · Jack Clark, "How Quickly Will A.I. Agents Rip Through the Economy?" (2026-02-24) · 2026 episode index (46 episodes Jan–Jul) · 2025 episode index (86 episodes)
• 2025 primary: "The Government Knows A.G.I. Is Coming" (2025-03-04) via LessWrong repost (2025-03-05) · Eliezer Yudkowsky episode (2025-10-15) · Kirk column (2025-09-11) via Volokh excerpt · Abundance (2025-03-18) · Goodreads ratings count
• Earlier primary: "This Changes Everything" (2023-03-12) via 3 Quarks Daily excerpt · 80,000 Hours — "Existential risk from AI and what DC could do about it" (2023-07-26) · "Democrats Have a Better Option Than Biden" (2024-02-16) · Vox founding (2014); Why We're Polarized (2020-01) — Wikipedia timeline
• Grading data (market-level): TechPolicy.Press — state AI legislation at mid-2026 (2026-07-06) · Roll Call — "At the Races: AI voted?" (2026-09-17) · NYT/Siena AI numbers (2026-09-16) · NYT/Siena sample and data-center numbers (2026-09-16) · Sanders–Casar Ban Artificial Superintelligence Act (2026-09-03) · Amodei, "We Must Pace the Frontier" (2026-09) · Naked Capitalism timeline of the September 2026 pacing debate (secondary, dates Amodei essay 2026-09-12) · Gary Marcus rebuttal (2025-03-05) · Biden withdrawal (2024-07-21)
• Incentive map: NYT v. Microsoft and OpenAI timeline (filed 2023-12-27; DOJ brief 2026-09-02; filings 2026-09-17) · NYT–Amazon AI licensing deal (2025-05-29) · NY Post via Yahoo — $40–70K speaking fees (2025-12-29) · Washington Speakers Bureau page (topics: Abundance, polarization) · NOTUS/Washington Sun — Build America Caucus (2025-05-08) · SF Standard — governor's housing forum (2026-05-08) · Axios — Senate Democrats retreat (2025-05-06; not fetched, 403) · Podtrac July 2026 ranker (2026-08-11)
• v1 sources carried: The Ezra Klein Show AI episodes with Altman, Amodei ("What if Dario Amodei Is Right About A.I.?", 2024-04-12), Hassabis, Yudkowsky; NYT AI columns 2023–present; Vox; Why We're Polarized (2020)

migrated v1→v2 2026-09-21 · backup: _archives/ezra-klein.html.bak-20260921 · converts-from: people/ezra-klein.md · AI & Society domain