Jasmine Sun

persona · active · confidence: medium · reviewed 2026-09-21 · tier: quarterly

Who she is

Affiliation: Independent writer and journalist, San Francisco — role check run 2026-09-21. She left Substack in 2025 after four years leading its core product team (her About page; Dialectic, 2026-06-19: "took a large pay cut," "a year of runway"). Today she wears six hats, five of which were missing from the May-2026 v1 file: (1) the newsletter at jasmi.news — tagline "an anthropology of disruption," over 38,000 subscribers on 2026-09-21; (2) contributing writer, The Atlantic (Dialectic bio, 2026-06-19; her Muck Rack bio reads "contributor @theatlantic"); (3) 2026 Omidyar Network Reporter in Residence — one of nine, announced 2026-01-15: "Over the course of their six month residences, each journalist receives a stipend, travel expenses, and access to Omidyar Network staff and partners"; (4) co-reporter on Kevin Roose's The AGI Chronicles (Farrar, Straus and Giroux, on sale 2026-10-06, "more than 150 interviews with insiders at OpenAI, Anthropic, and Google DeepMind") — her words, 2025-12-31: "I worked with Kevin Roose to report a history book on the AGI race"; the publisher page does not name her and the book's own site refused connections on 2026-09-21, so this credit rests on her statement until the book ships; (5) co-founder (2020) and co-editor-in-chief of Reboot, "a publication by and for technologists," and its annual print magazine Kernel (issue 6, "FEED," on sale July 2026); (6) keynote speaker represented by Leigh Bureau. Earlier: AI-policy research at Mozilla ("Public AI" white paper, 2024), Schmidt Futures and the International Climate Development Institute in Taipei; a Day One Project policy memo (2020); an Emergent Ventures grant, 39th cohort, 2024-12-18 — "Jasmine Sun, San Francisco, for full-time writing and Substack." Stanford BA in sociology. Bylines on her portfolio page: New York Times (Sunday Business, January 2026; Sunday Opinion, 2026-04-30), The Atlantic (March 2026), WSJ Ideas (2025), SF Standard (2025), Business Insider (2025), Substack Reads (2025), NPR segments (2025–26). No Rest of World byline exists — see the Wrong grade below.
One-line position: The v1 line — AI debates miss the social, generational and political contexts the technology lands in; the meaningful story is cultural, not just technical, and Silicon Valley's self-narration deserves close reading from inside the milieu — still describes the method. The 2026 position is sharper and more economic: the insiders' "quiet consensus" is that "in the next few years, we are going to see mass job displacement" (80,000 Hours, 2026-06-04); "Most people I know in the A.I. industry think the median person is screwed, and they have no idea what to do about it" (NYT Opinion, 2026-04-30, via her 2026-05-13 companion post); and the public's answer is AI populism — "All the money is on one side and all the people are on the other" (2026-02-11) — which she reports from inside the industry's parties and from the Wisconsin towns fighting its data centers.

Discipline & technical bet

Sociology by training, product management by trade (four years on Substack's core product — community features, publishing, podcast/video tools), reporting by choice since 2025. Her stated method is ethnographic: "long-form AI ethnographies" built on off-the-record conversations with researchers (80,000 Hours, 2026-06-04); "Few journalists actually live in San Francisco, which Jasmine finds absurd" (Dialectic, 2026-06-19); and Sontag's rule that "the critic's first duty is to explain, not to evaluate." She is not building anything, but she does carry two concrete technical wagers. (a) Writing quality is bottlenecked by evaluation and business priority in post-training, not by model scale — from her Atlantic essay (March 2026) and its 2026-03-23 companion: "Good writing is hard to train because it's hard to evaluate"; "Good writing is not a business priority, so other post-training goals get in the way"; "Good writing requires a grounding in real-life experience"; "The factory settings are designed for a sycophantic corporate assistant, not a creative genius." Editing "is closer to a role that LLMs can play." (b) High-agency coding agents are the real capability shift — after a week with Claude Code (2026-01-23): "It is an incredible technology that has made me more AGI-pilled than ever, while also being a net decrease on my work productivity"; on agents doing whole professional tasks: "Not yet. But probably in a year." A related product bet: "It will soon cost near-nothing to have whatever app you want" — a "home-cooked app renaissance" (the thesis of her WSJ Ideas piece, "Your Next Favorite App? The One You Make Yourself," 2025). Layer focus: presentation — how AI is felt, used and fought over — with 2026 reporting reaching into infrastructure through the data-center backlash.

Key claims (Says)

Notable predictions — with falsifiable checks

Revealed behavior (Does)

Feels

Fear that the people she lives among are right about the median person and wrong about what to do — her nightmare is havoc without cures. Fear of self-deception as the AI-age vice: "AI makes it too easy to lie to yourself" (2026-06-10); "The only question AI can't answer for you: will you lie to yourself?" (Dialectic). Hope that writing survives because it "makes her smarter and kinder," and that "now is basically the most exciting time in the entire world" for independent writers — a hope her own income depends on.

Hears

Kevin Roose and the NYT Opinion desk; Ezra Klein; the Substack ecosystem she helped build; the Reboot circle (Jessica Dai, Jacob Kuppermann, Hamidah Oderinwale); China watchers (Sinica, ChinaTalk's Irene Zhang); Sontag on criticism; Karnofsky and Christiano on risk (her recommended alternative to Yudkowsky); named interlocutors in the populism piece — Dean W. Ball, Andy Masley, Max Tegmark, Josh Hawley, Richard Blumenthal; Benjamín Labatut on literature; and, off the record, the lab researchers whose "quiet consensus" she reports.

Sees

Three vantages most commentators lack: the inside of the SF social world (parties, house dinners, off-record researchers — "Few journalists actually live in San Francisco"); the inside of the labs' paper trail through 150+ Roose-book interviews and "unpublished company documents"; and the outside — Midwestern town halls, recall campaigns, NDAs, and Chinese workers on 3,000-RMB salaries. Plus a builder's eye: she ran Substack's core product and spent "every day last week talking to Claude Code more than my friends."

Incentive map

Re-derived from scratch this cycle; v1's "not financially exposed to a specific AI thesis" was graded on equity alone and is Drifted. Who pays her: (1) paid newsletter subscribers — 38,000+ total subscribers on 2026-09-21; "over 10,000 free subscriptions" gained in 2025 plus a Substack "Bestseller badge," which implies a paying tier (price not displayed on the subscribe page as fetched; funnel arithmetic not possible this cycle); (2) The Atlantic, as a contributing writer, and freelance features at the NYT, WSJ, SF Standard, Business Insider; (3) the Omidyar Network residency (stipend, travel, "access to Omidyar Network staff and partners," six months from early 2026); (4) the Roose book contract — one of the "contracts that were my main income sources" in 2025; (5) Leigh Bureau speaking fees; (6) Emergent Ventures (Tyler Cowen's Mercatus program) — twice: Reboot lists it as a funder alongside Omidyar Network, Ford Foundation, RT Youth Power Fund, 0xPARC and SPARC, and she received a personal grant in December 2024. Reboot itself pays her nothing; it is a fiscally sponsored, volunteer-run nonprofit. What she is selling: access-plus-distance — the insider who will tell you what researchers say off the record, and the reporter who will drive to Janesville. Both halves are the product. Funder–thesis pairing (log it): her newest vehicle (Omidyar's rules-institutions-norms residency) matches her newest thesis (AI populism as the story of 2026); the thesis may be sincere and still be the deliverable. What she can't say: that the SF insider circle isn't worth watching (her whole franchise); that Substack's model fails writers (former employer, host platform, Substack Reads feature, Substack Media Forum talk 2026); anything that burns off-record lab sources before the book ships; and, until 2026-10-06, anything that pre-empts Roose. Counter-evidence to capture: the underclass piece and the Yudkowsky pan both cost her something with a pole. Rule-25 discount: moderate — no lab or Substack equity found; the live conflicts are access-dependence, funder-topic alignment, and the Cowen link that v1's "splits sharply on politics" line never mentioned.

Theories aligned with

What she's reacting against

Where she overlaps / splits (with Rich)

Track record

Empirical vs normative

Weak spots / open questions

Rich's take

Delta log

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

Sources

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