Ed Zitron

Who he is

Affiliation: CEO, EZPR (media-relations firm, founded 2013) — role check run 2026-09-21: unchanged per his About page. Writes Where's Your Ed At (2020–; About page now claims 105,000 subscribers, vs 50K+ in Feb 2025 per Wikipedia; paid tier $7/mo, $18/qtr, $70/yr). Hosts Better Offline (Cool Zone Media / iHeartPodcasts, 2023–; Webby winner; 272 episodes by 2026-09-15). Correction to v1: the podcast is Better Offline, not "Where's Your Ed At audio". Book Why Everything Stopped Working (Penguin Portfolio) — About page now says mid-2027; 2025-era sources said 2026. Born 1986/87, Hammersmith; New York and Las Vegas.
One-line position (v1, carried): The frontier AI buildout is a bubble — the economics don't work, the products don't deliver on the promises, and the tech industry is engaged in a collective delusion sustained by hype, cheap capital, and regulatory capture. 2026 narrowing: "the AI bubble IS the OpenAI bubble" — collapse "likely around 2027" when OpenAI's funding needs meet tightening credit (The OpenAI Bubble, 2026-07-15). Since 2026-06-15 he holds primary documents: OpenAI's audited 2024–25 financials, verified by the Financial Times.

Discipline & technical bet

Non-technical: games journalist → PR executive → newsletter polemicist. No technical wager; the bet is financial-forensic. 2026 form: the whole AI trade is one credit event wearing many hats — OpenAI plus Anthropic are ~70% of hyperscaler AI revenue, and 2027 is the "reset wall" when $200B+/yr of compute contracts come due (Concentration Risk, 2026-09-08). Method: filing footnotes (RPO = revenue promised but not yet delivered), leaked documents, biggest number divided by smallest. Layer: infra. Different animal from Marcus: Marcus says the models can't, Zitron says the money can't — and this cycle he quietly conceded the capability axis: "LLMs… have improved in the last year" (2026-09-01); code debugging works acceptably (2026-08-25).

Key claims (Says)

• Empirical: The AI capex-to-revenue gap is unsustainable — hundreds of billions in infrastructure spending not matched by proportional revenue; "a structural problem, not a timing issue" (v1). Drifted — mechanism held: 2026 hyperscaler capex ~$725B, +77% (TMT Finance); Alphabet Q2 2026 free cash flow −$5.86B on $44.9B quarterly capex (8-K, 2026-07-22); S&P cut Oracle to BBB− naming OpenAI a "central credit risk" (2026-07-09). "Not timing" failed: NVIDIA data-center revenue $89.0B, +117% YoY (2026-08-26); Anthropic run-rate $65B, 7x since end-2025 (CNBC, 2026-08-17); OpenAI $40B, ~2x (Bloomberg, 2026-08-13). He has re-dated the reckoning three times — a timing claim by his own behavior — and now says "I underestimated the sheer desperation" (2026-08-25).
• Empirical: AI products mostly don't work well — Copilot adoption is overstated, AI Search is worse than traditional search, enterprise AI deployments are underperforming; the demos don't translate to daily use (v1). Wrong (strong form) — retired by its own author. M365 Copilot passed 30M paid seats, net adds more than doubling QoQ (Nadella, 2026-07-29); OpenAI coding/work products 20M weekly users (Bloomberg, 2026-08-13). His defensible remainder — paid Copilot is ~6.6% of paid commercial seats (Directions on Microsoft) — is a penetration argument, not "doesn't work". He conceded it himself: adoption acknowledged, LLMs "function acceptably" for narrow supervised tasks (2026-08-25). Residual "no killer app": check below.
• Empirical: This is a bubble — in the specific economic sense: investment has outpaced revenue and will correct; the parallel is dot-com, not electricity (v1). Wrong on every dated version to date; the undated core is unfalsifiable and is replaced by the 2027 check below. NVIDIA market cap $5.37T on 2026-09-18, a record (StockAnalysis); the 2026-09-14 selloff (Nasdaq −1 to −1.3%, chip index −5.9%) was triggered by an AI-safety slowdown call, not economics (Bloomberg, 2026-09-13). Valence note: dot-com support survives at the edges — Alphabet negative FCF, Oracle at BBB−, SoftBank on junk bonds.
• Normative / mixed: Tech media is complicit — journalists and analysts amplify corporate claims without scrutiny, creating a feedback loop of hype (v1). Unfalsifiable — rewrite or retire. The observable now cuts against him: the FT verified his own leaked documents and Fortune, Bloomberg, Forbes carry him as a source. Rewrite: "no major outlet publishes primary-document AI-economics scrutiny without a leak" — checkable yearly.
• Empirical: Google Search is getting worse — the "Rot Economy" thesis: AI integration has degraded core products, not improved them (v1). Drifted — mechanism held, valence wrong. Mechanism (growth-chasing degrades the product) held on the pre-AI record: DOJ trial documents, Leipzig SEO-spam study, Raghavan out within six months (Sloppish scorecard, 2026-03-20 — adversarial, still marks this correct). Valence (AI integration hurts the product or business) wrong on market data: Search & other +17% to $63.3B in Q2 2026, AI Overviews past 1B monthly users (Alphabet 8-K, 2026-07-22). No market-level quality metric exists; noted, not smoothed over.
• Meta-claim: The "AI will be bigger" argument is unfalsifiable — proponents always say returns are "just around the corner"; this is the structure of a bubble narrative (v1). Unfalsifiable as stated. Reversal noted: the bull side's dated numbers have hit (Anthropic 7x, Copilot 30M) while his have missed (Luu: 28 of 29). Rewrite: name one bull prediction and one date.
• Empirical, NEW 2026: OpenAI's audited 2025 — revenue $13.07B, costs $34B, operating loss $20.92B, net loss attributable $38.53B (2024: $5.09B); $17.2B paid to Microsoft; $867M of revenue from SoftBank, $303M from Microsoft; ~$50B assets, about half cash. Held — "independently verified by the Financial Times" (exclusive, 2026-06-15; Fortune, 2026-06-16). Open: how much of the $38.5B is non-cash — his own reading shows $17.87B + $3.95B of noncontrolling-interest reclassification.
• Empirical, NEW 2026: Demand concentration — OpenAI + Anthropic ≈ 70% of hyperscaler AI revenue; OpenAI alone over 7% of Microsoft's total revenue. Held on the filing — Microsoft's FY2026 10-K: $24.1B from OpenAI (7.26% of $331.8B), $6.0B receivable (10-K; his read, 2026-08-05). The "70% of AI revenue" is his derivation from Microsoft's stated AI run-rate — an inference. NVIDIA: five direct customers = 70% of receivables (22/14/13/11/10%) at 2026-07-26 (10-Q); his "44% of FY27 revenue from three customers" was not found in the filing narrative — unverified.
• Empirical, NEW 2026: The AI bubble is the OpenAI bubble — Oracle's ~$300B OpenAI contract ≈ half its RPO; SoftBank's $40B bridge; IPO slipping. Held (mechanism, live) — S&P's downgrade named OpenAI and forecast a ~$42B Oracle FY27 free-operating-cash-flow deficit; Oracle Q1 FY27 capex $28.5B, RPO $664B (CNBC, 2026-09-10); Altman: a 2026 IPO "ill-advised", now 2027 (Axios, 2026-09-12). Unabsorbed counter-evidence: SoftBank repaid the $25.9B bridge balance early (2026-09-15) and is placing up to $20B of bonds — the credit channel he expected to tighten is still open (Bloomberg, 2026-09-09).
• Empirical, 2024→2026: Subsidized pricing must reprice; costs to serve rise (Subprime AI Crisis, 2024-09-16; "AI Is Getting Way Too Expensive", 2026-07-31). Drifted — per-token prices fell ~99% since GPT-4 launched (Piper, 2026-04-28), so the per-unit version is wrong; total cost to serve did rise ($34B on $13.07B — his own scoop) and token-billing repricing did happen (Cursor, 2025). Garnish error on record: Cursor "$5,000/month per user" vs ~$400, about 12x (Sloppish fact-check).

Notable predictions — with falsifiable checks

• (2024) AI revenue would not justify capex levels — v1 outcome: partially supported; hyperscaler revenue is growing but capex growth is faster. Drifted — the dollar gap widened (capex ~$725B vs combined lab run-rates ~$105B) while revenue growth rates are the fastest on record. Check: hyperscaler AI revenue growing faster than AI capex for two consecutive quarters by 2027-12-31 falsifies him; a big-four capex-guidance cut supports him.
• (2024) Google Search quality would continue declining with AI integration — v1 outcome: contested; AI Overviews received criticism but search revenue held. Drifted — still no quality metric; revenue +17%. Check: Search & other growth under 10% YoY for two quarters by 2027-12-31, or DOJ-remedy-era query data showing decline.
• (2024-09-16) "Whether or not it's Q3 or Q4 2024, or even Q1 2025, Wall Street will begin punishing big tech" (Subprime AI Crisis). Wrong — NVIDIA $5T (Oct 2025), $5.37T (2026-09-18).
• (2024) OpenAI dead by the end of 2025. Wrong — self-scored: "I said OpenAI would be dead by the end of 2025 back in 2024… I underestimated the sheer desperation" (Manifesto, 2026-08-25). His first public scoring of a miss — a delta against the investing twin and Sloppish, which had found none.
• (2025-10) The bubble pops "no later than Q2 2026". Wrong — "no later than" made it an upper bound (Luu, Sep 2026); Q2 ended with records.
• (2026-07-15 / 2026-08-18) Collapse "likely around 2027"; OpenAI's failure would be "the Lehman Brothers of the AI bubble"; OpenAI needs "tens of billions" before end-2026 and another round by March 2027 (2026-08-18). Live. Checks: (a) OpenAI closes ≥$30B new primary capital by 2027-03-31 — mechanism damaged; (b) OpenAI misses or renegotiates a cloud payment, or Oracle books an OpenAI-related impairment, by 2027-12-31 — confirmed; (c) NVIDIA data-center growth under 20% YoY in any FY2028 quarter — his own "slow down meaningfully" trigger fires; (d) any big-four capex-guidance cut — supports.
• (2026-08-18) His own falsification bar: "For me to be wrong, in the space of three years OpenAI will have to become a company with annual revenues higher than Meta ($200 billion)". Live — but a ratchet: from "dead by 2025" to "must out-earn Meta by 2029". Softer test he should accept: OpenAI at run-rate operating breakeven by 2028-12-31 guts the thesis without $200B.
• (Ongoing, v1) The AI bubble will correct — outcome: TBD; the central falsifiable claim; no major correction as of mid-2026. Unfalsifiable without a date; superseded by the 2027 item.
• (2026, residual) "No killer app". Check: any single AI product line with disclosed revenue above $10B/yr by 2027-12-31 falsifies it.

Revealed behavior (Does)

• Where the time went since the last review: 24 posts from 2026-06-26 to 2026-09-18, twelve of them paid "Hater's Guides" (NVIDIA, Oracle, SoftBank, Memory, Broadcom, Circular Financing ×2, AI Debt) — about half the output is paywalled; the paid tier is the product (archive).
• Published FT-verified audited OpenAI financials (2026-06-15) and a filing-based read of Microsoft's FY2026 10-K (2026-08-05) — commentary became sourced reporting, which v1 said he lacked.
• Hedged for the first time, twelve days after Luu's 28-of-29 audit: "You'll notice I'm a little light on predictions, and that's because everything is a little volatile right now" (2026-09-14). No direct response to Luu found as of 2026-09-21.
• Same piece pivoted to present-tense harm (ChatGPT and suicides; the OpenAI–Hugging Face sandbox-escape) and remedies (prosecution, halt training) — a new axis outside his economics lane.
• Doubled the newsletter to 105K, slid the book to mid-2027, kept EZPR. No investment positions disclosed; none found.

Feels

Wants it to be true that "the most devout defenders of AI could become great critics"; fears the sunk-cost spiral where "every step into madness will be more embarrassing to explain" (2026-08-25) — the trap his own timeline ratchet exhibits. The engine is moral outrage at manufactured narrative, from a man whose day job is manufacturing narrative.

Hears

Filings and footnotes; leaked documents via a Signal tip line; FT reporters as verifiers; Cool Zone Media; Brian Merchant, Doctorow (enshittification is the Rot Economy's sibling); Covello/Cahn capex math; a 105K-reader reply audience. Almost no technical input — no ML researchers, no benchmark literature — which is why capability claims are his weakest lane.

Sees

Documents reporters don't get (audited financials from a company that publishes none); PR from the inside — how a narrative is assembled and sold; the RPO and credit footnotes most tech press skips; a reader base large enough to be a tip network.

Incentive map

Re-derived this cycle and diffed against the --investing twin (reviewed 2026-07-19, not modified). The twin had: PR CEO; a subscription brand "structurally rewarded for doom"; no disclosed positions; a book forthcoming. Diff: (1) audience roughly doubled to 105K (About page) from 50K+ (Feb 2025) — at $70/yr each 1,000 paid readers is $70K/yr; paid rate undisclosed, so no revenue figure is asserted. (2) About half the output is now paywalled "Hater's Guides" — doom is the SKU. (3) The book slid to mid-2027, the year of his predicted collapse; a bubble that fails to pop before publication is a commercial problem, not only an intellectual one. (4) Sloppish alleges EZPR has AI clients; the client list is not public — unverified, flagged. (5) No short positions or investments disclosed; searched 2026-09-21, none found. (6) Mainstream validation (FT, Fortune, Bloomberg) raises the cost of a 2027 miss — fewer notches left on the ratchet. (7) What he cannot say: "AI is useful and the buildout is merely oversized" — Cahn's and Covello's lane, and it does not sell a $70 subscription. Rule-25 discount: high on timelines and terminal claims; low on document-based facts, which have verified.

Theories aligned with

• Technology-bubble economics (capex-vs-revenue analysis) — v1
• Media criticism / tech-industry accountability — v1
• Adjacent to Marcus's capability skepticism but from an economics/media frame rather than a technical one — v1
• Counter-voice to techno-optimism — the bench's designated hostile witness
• NEW: circular-financing / vendor-financing critique (NVIDIA→CoreWeave→NVIDIA; SoftBank inside OpenAI's revenue line) — shared with Tooze
• NEW (2026-09): AI-harm and prosecution framing — a normative turn, one post old

What he's reacting against

• AI industry hype cycle — argues most AI revenue claims are vaporware and the capex-to-revenue ratio is unsustainable (v1)
• Mainstream tech media that amplifies corporate AI narratives uncritically — positions himself as the contrarian voice calling out the emperor's new clothes (v1)
• The assumption that AI products are delivering real value — insists the user experience and economic returns don't justify the investment (v1)
• Silicon Valley's cultural inability to acknowledge failure — argues the industry has substituted narrative for evidence (v1). Irony logged: 28 misses of 29, none acknowledged until 2026-08-25
• NEW: "hypernormalization" — circular deals and perception management standing in for value (2026-09-01)
• NEW: bailouts — "No bailouts, no handouts, no special treatment, no tax breaks, no CHIPS act, and no sovereign wealth fund" (Let AI Burn, 2026-07-07, per the twin)
• NEW: x-risk discourse as a distraction from present harms (2026-09-14) — puts him, unusually, against Amodei and the safety camp too

Where he overlaps / splits (with Rich)

• Overlaps with Covello (Goldman Sachs) on capex-revenue skepticism; splits on register — Covello writes institutional research, Zitron writes provocative newsletter essays (v1)
• Overlaps with Cahn (Sequoia) on the revenue gap; splits on tone — Cahn framed it as a question ("$600B Question"), Zitron treats it as an answer ("it's a bubble") (v1)
• Overlaps with Marcus on AI products not delivering; splits on focus — Marcus questions capabilities, Zitron questions economics (v1); Zitron has now conceded capability ground Marcus has not
• Overlaps with Tooze on macro-financial skepticism; splits on depth — Tooze places it in historical context, Zitron writes polemic (v1)
• Splits with Huang / Nadella / Pichai fundamentally — they say the investment is rational, Zitron says it's delusional (v1); Nadella's 30M seats and Pichai's +17% Search are this cycle's market-level rebuttals
• Overlaps with Aschenbrenner on the trigger (capex deceleration, 2026–27); splits on terminal value — repricing inside a real bottleneck vs stranded assets (from the twin)
• Overlaps with Evans on distrusting headline numbers; splits on what follows — Evans says they can't be scored, Zitron scores them as fraud; Luu scored Zitron instead
• With Rich: overlap is the two-lab stress test — if ~70% of hyperscaler AI revenue is two subsidized labs, every picks-and-shovels position is exposed to one credit event; split is that Rich reads scarcity as pricing power, Zitron as fragility. Let him name the failure mode, never size the position (twin's protocol)

Track record

• "Rot Economy" thesis gained significant traction — genuine influence on how people think about tech product quality degradation (v1). Held — even the adversarial Sloppish scorecard marks it correct
• Consistently provocative and attention-getting — large newsletter audience for a solo voice (v1). Held — 105K, and a source of record after the OpenAI scoop
• Specific claims about product failure (Copilot, AI Search) are sometimes exaggerated relative to measured adoption data (v1). Confirmed harder: 30M paid seats; Luu's audit
• Has been calling "bubble" for over a year without correction occurring — the timing problem of all bubble callers (v1). Now two-plus years and quantified: 28 of 29 dated 2024–25 predictions wrong (Luu, Sep 2026); Sloppish: 3 right, all pre-AI (Rot Economy, Google Search, remote work), 6 wrong
• Landed: "The Man Who Killed Google Search" (2024-04, DOJ documents); Subprime repricing mechanism (partial); the OpenAI scoop (FT-verified); the Microsoft 10-K read; Oracle downgraded on the channel he had named
• Missed: Meta "a dying company" (2022; Nov 2024); Q3 2024–Q1 2025 punishment; OpenAI dead by 2025; Q2 2026 pop; inference costs won't fall; the ~12x Cursor number
• Pattern: elite at forensic incentive analysis and cost-side accounting; persistently wrong on timing and terminal outcomes; first self-scored miss 2026-08-25, first hedge 2026-09-14

Empirical vs normative

• Empirical: capex data, revenue comparisons, product adoption analysis, search quality metrics (v1); now also audited financials, 10-K related-party disclosures, rating actions
• Normative: the tech industry should be held accountable; media should scrutinize AI claims; investment should be proportional to evidence (v1); 2026 additions: no bailouts, criminal liability for harms, halt training
• The empirical work is genuine but selectively presented — he systematically highlights negative evidence and minimizes positive signals (v1). Still true: Anthropic's $65B is metabolized as "Anthropic overtakes OpenAI", not "demand exists"
• Conflict note (CONVENTIONS rule 25): subscription-plus-book business whose revenue rises with doom, plus a PR firm with an undisclosed client list — high discount on forward claims; his verified documents carry almost none

Weak spots / open questions

• Bubble-calling has a timing problem — you can be directionally right but wrong on when, and being early is indistinguishable from being wrong until the correction happens (v1). Now quantified at 28 of 29
• Selection bias is the central concern: he systematically presents the worst case and ignores counter-evidence (Azure AI revenue growth, enterprise adoption data) (v1). This cycle's omissions: Azure past $100B, Google Cloud +82%, SoftBank's early repayment
• Not a technologist or economist — the analysis is commentary, not research; no peer review, no formal methodology (v1). Partly cured by the documents; not on capability claims
• PR firm CEO critiquing tech industry hype creates its own conflict of interest — contrarianism is a brand positioning, not just an analysis (v1). Sloppish's AI-clients allegation is unverified; Rich to decide whether it matters
• If the AI buildout produces genuine economic returns (even modest ones), his entire frame is wrong — and the evidence for modest returns is growing alongside the evidence for overinvestment (v1). Both grew again this quarter
• NEW: the $200B-in-three-years falsification bar is a ratchet — hold him to the breakeven test
• NEW: "44% of NVIDIA FY27 revenue from three customers" not found in the 10-Q narrative — his number until located
• NEW: the harm/prosecution turn (2026-09-14) is normative, outside his evidentiary lane, and risks the source-of-record status the scoop earned
• NEW: who leaked audited financials to a PR-firm CEO, and why (twin) — open

Rich's take

• (your synthesis here)

Delta log

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

• Grades: 4 Held / 5 Drifted / 5 Wrong (2 unfalsifiable items marked for rewrite, not graded; 2 live 2027 predictions with checks attached). Review window 2026-06-16 → 2026-09-21.
• Wrong (5): "AI products mostly don't work" — 30M paid Copilot seats (2026-07-29), his own concession (2026-08-25); "this is a bubble that will correct" on every dated version — Q3 2024–Q1 2025 (Subprime AI Crisis, 2024-09-16), OpenAI dead by 2025 (self-scored 2026-08-25), "no later than Q2 2026" (Luu, Sep 2026); NVIDIA record $5.37T on 2026-09-18.
• Drifted (5): capex-revenue gap "structural not timing" (gap held, timing failed, re-dated three times); Google Search worse (mechanism held on the DOJ record, valence wrong on +17% Search revenue, Q2 2026); costs-must-reprice (per-token −99%, total cost to serve up); the two 2024 predictions on capex justification and Search decline.
• Held (4): OpenAI audited 2025 figures (FT-verified, 2026-06-15); demand concentration (Microsoft 10-K: $24.1B from OpenAI, 7.26% of revenue, filed Aug 2026); "the AI bubble is the OpenAI bubble" transmission map, mechanism only (S&P Oracle to BBB− naming OpenAI, 2026-07-09; IPO to 2027, 2026-09-12) — terminal outcome live; Rot Economy / product-degradation record.
• Most surprising delta: the bench's least self-correcting voice scored his own miss ("OpenAI would be dead by the end of 2025… I underestimated the sheer desperation", 2026-08-25) and, twelve days after Luu's 28-of-29 audit, published his first hedge (2026-09-14). The twin's "has never publicly scored his own misses" is now stale.
• Employer / incentive check (mandatory): no job change — EZPR CEO, unchanged (About page, read 2026-09-21). Incentive map re-derived and diffed against the twin (see section). Dated post-review sources: his posts 2026-06-26 → 2026-09-18; Luu (Sep 2026); Microsoft 10-K (Aug 2026); NVIDIA (2026-08-26); Altman (2026-09-12); SoftBank (2026-09-09).
• v1 corrections: podcast is Better Offline, not "Where's Your Ed At audio"; "no formal methodology" superseded by FT-verified documents.
• Tier: quarterly — his 2027 window opens in Q1 2027 (OpenAI raise by March, SoftBank bond placement, Oracle cash flow); each quarter until then can grade the central prediction. Confidence unchanged at medium: the documents verify, the dates don't.

Sources

• His own, post-review (primary): OpenAI audited financials exclusive (2026-06-15) · The OpenAI Bubble (2026-07-15) · Microsoft disclosures (2026-08-05) · What Happens If OpenAI Dies? (2026-08-18) · AI Hater's Manifesto (2026-08-25) · Hyperscale Normalization (2026-09-01) · Concentration Risk (2026-09-08) · AI Is Already In Dangerous Hands (2026-09-14) · archive · About
• His own, earlier: The Subprime AI Crisis (2024-09-16) · "The Rot Economy" (2023-02) · "The Man Who Killed Google Search" (2024-04) · The Hater's Guide To The AI Bubble (2025-07)
• Adversarial reads (mandatory pairing): Dan Luu — 29 predictions scored (Sep 2026; HN 2026-09-02) · Sloppish scorecard (2026-03-20) · Sloppish fact-check · Piper, The Argument (2026-04-28)
• Market-level data used for grades: NVIDIA Q2 FY27 release (2026-08-26) · NVIDIA 10-Q, quarter ended 2026-07-26 · Alphabet Q2 2026 8-K (2026-07-22) · Microsoft FY2026 10-K (Aug 2026) · Copilot 30M paid seats (2026-07-29) · Directions on Microsoft — Copilot penetration · Anthropic $65B run-rate (CNBC, 2026-08-17) · OpenAI $40B run-rate (Bloomberg, 2026-08-13) · OpenAI $122B round at $852B (CNBC, 2026-03-31) · S&P cuts Oracle to BBB− (2026-07-09) · Oracle Q1 FY27 (CNBC, 2026-09-10) · SoftBank repays bridge early (Bloomberg, 2026-09-09) · OpenAI IPO to 2027 (Axios, 2026-09-12) · 2026-09-14 selloff (Bloomberg) · NVIDIA market cap history · TMT Finance 2026 capex analysis · Reuters — Blue Owl exits Oracle's Michigan project (2025-12-17), the trigger the twin lists
• Bio: Wikipedia · Fortune on the scoop (2026-06-16) · Better Offline (iHeart)
• v1 sources (carried): Newsletter Where's Your Ed At · Podcast (v1 said "Where's Your Ed At audio"; corrected to Better Offline) · Media: widely cited in mainstream tech journalism as a contrarian voice · "Rot Economy" essay: the foundational piece on tech product quality degradation

migrated v1→v2 2026-09-21 · backup: _archives/ed-zitron.html.bak-20260921 · converts-from: people/ed-zitron.md · AI & Society domain · investing twin: ed-zitron--investing.html (reviewed 2026-07-19, read, not modified)