Who he is
Affiliation: George Mason University (Holbert L. Harris Chair of Economics); Mercatus Center; Marginal Revolution blog; Conversations with Tyler podcast; column now at The Free Press (moved from Bloomberg). Verified current as of Aug 2026 — no job change.
One-line position: AI is real and big — capability may already be "AGI on most things humans care about" — but diffusion is slow, gains are uneven, and institutions are the bottleneck.
Discipline & technical bet
Generalist economist whose method is volume and synthesis: read everything, talk to everyone, update daily in public. The concrete wager, sharpened 2025–26: capability is essentially solved ("we have AGI already" on most real human tasks) but the binding constraint is human and institutional absorption — "the #1 bottleneck to AI progress is humans." He is long capability, short diffusion.
Key claims (Says)
- AI is a major GPT but society absorbs change slowly — regulation, status, org friction. Held — July 2025 DeepMind discussion doubles down: rent (regulatory), food (decades), healthcare (20–40 years), education (uncertain).
- O-ring effects: AI lifts the most capable users disproportionately; the median worker doesn't see big income gains. Held (early evidence) — Canaries payroll data (Aug 2026): experienced workers keep employment and gains while entry-level contracts; the experience premium is the o-ring signature.
- AI accelerates science faster than bureaucracy. Held — Olympiad math solved (2024), AI-designed drugs in human trials (Isomorphic, 2026), while healthcare delivery and education barely move — the asymmetry he predicted.
- Human capital becomes more valuable, not less — agency, taste, judgment scarcer. Held — same experience-premium evidence; his status-redistribution warning (below) is its political corollary.
- Macro growth lift modest near term, visible ~2030. Drifted — arriving earlier than his marker: labor productivity ~2.5%/yr annualized 2022Q3–2025Q2 (KC Fed, Feb 2026) and nominal AI GDP ~$250B in 2025 with ~2,600%/yr quality-adjusted growth — numbers he himself now amplifies while urging satellite accounts.
- Geography matters: US > Europe > most of world in deployment. Held — no counter-evidence; his talks increasingly emphasize US-China bipolarity.
- NEW (Apr–Jul 2025): "we're at AGI on most things real humans care about" (declared o3 AGI, Apr 2025) — capability question closed for him; only diffusion remains.
- NEW (2025–26): status-redistribution risk — "rapid redistribution of status away from groups that can make a big stink about it" (educated professionals) may create political instability.
- NEW (May 2026): the measurement gap — national accounts weren't built for AI; statistics agencies need AI satellite accounts before the measurement gap becomes a policy gap.
Notable predictions — with falsifiable checks
- (2023) "AI changes everything but slowly" — 10–20 year diffusion. Held so far — but now in open tension with his own AGI-now claim. Check: share of US firms reporting material AI process change (Census BTOS) crossing 50% — before or after 2030?
- (2024) US growth lift visible by ~2030, not before. Drifted — sectoral productivity acceleration and an AI production boom are visible in 2025–26 data. Check: BEA/BLS aggregate TFP ≥0.5pp above 2015–19 trend for a full year before 2030 = early; never by 2032 = late.
- (2023) AI widens the high-agency / low-agency gap. Held (early) — check: wage/employment dispersion by experience tier in exposed occupations through 2028.
- (2024) Healthcare and education slowest to absorb gains. Held — check: sector productivity contributions still bottom-quartile in 2028 data.
- (Multiple) Measurable AI science acceleration in specific fields within 5 years. On track — math done; biology's verdict is the Isomorphic-class clinical readouts ~2028.
Revealed behavior (Does)
- Bet his own corpus on AI: published The Marginal Revolution: Rise and Decline, and the Pending AI Revolution (Mar 2026) alongside a "generative book" edition — he ships experiments, not just takes.
- Declared o3 AGI (Apr 2025) and took the mockery — a real reputational stake, not hedged.
- Spends time inside the labs (DeepMind London, July 2025) and with the builders (Dwarkesh, 4th appearance) — his input diet is the frontier, not the seminar room.
- Moved his column to The Free Press — audience-building over institutional prestige.
- Still blogs daily — the revealed belief that human curation retains value in an AGI world.
Feels
Wants the Great Stagnation to end on his watch and to be the one who called it correctly from both sides — fast capability, slow diffusion. Fears mood affiliation more than being wrong; visibly enjoys holding the AGI-now + slow-growth tension nobody else will hold.
Hears
Frontier lab researchers, Dwarkesh-circle builders, his MR comment section, and centuries of economic history — the widest input funnel in the bench, filtered through public-choice priors.
Sees
Cross-domain pattern matches faster than specialists: he connects capability news, GDP accounting, and status politics in the same week. Blind side: he samples elite users and builders; median-worker experience reaches him as data, not contact.
Incentive map
Paid by GMU/Mercatus (donor-friendly market-liberal frame); sells books, the blog's centrality, and access to everyone. The AGI-now call and the slow-diffusion call together are unfalsifiable as a pair in the short run — each escape-hatches the other; that's the incentive-shaped ambiguity to watch. Can't easily say: that regulation might be the right answer to something.
Theories aligned with
- Great Stagnation (AI as potential exit ramp)
- O-ring production theory (Kremer)
- Public choice / regulatory friction lens
What he's reacting against
- Doomerism (Yudkowsky-style x-risk framing) — treats it as a small but nonzero tail
- Hype cycles that confuse capability demos with economic deployment
- Status-quo bias dressed as moderation — also rejects "nothing will change"
- Mood affiliation in AI discourse — picking a side because of tribe, not evidence
- NEW (2026): benchmark fetishism — benchmarks bias toward what AI can't do; real-task performance says AGI is here
Where he overlaps / splits (with Rich)
- Overlaps with Brynjolfsson on slow diffusion; splits on frame — societal absorption vs org-level complements. Both now watch the same Canaries data.
- Overlaps with Acemoglu on modest near-term TFP; splits on cause — regulation/status friction vs market power and design. Their status-instability warnings (his "big stink" groups, Acemoglu's Gen Z revolution) are converging from opposite priors.
- Splits with Andreessen on ideological accelerationism while staying structurally optimistic
- Splits with Yudkowsky on x-risk magnitude
- Splits with Amodei on what capability means — Cowen says AGI is here and doesn't much matter yet; Amodei says it's coming and changes everything
- For Rich's bench: his AI-GDP/satellite-accounts thread is directly useful for the picks-and-shovels thesis — $250B nominal AI GDP growing ~2,600%/yr quality-adjusted is the revenue-side check on the capex story. His o-ring frame maps onto Rich's judgment-over-execution principle.
Track record
- Great Stagnation (2011): held through the 2010s; the 2025–26 productivity acceleration is the live test of the exit ramp
- Average Is Over (2013): predicted AI-complement bifurcation — the Canaries experience-premium data is its strongest confirmation yet
- o3-AGI call (Apr 2025): reputationally brave; verdict depends on whose definition wins
- Explicitly tracks his own record on the blog — self-correction machinery real, though the paired-claims structure (see incentive map) blunts it
Empirical vs normative
- Mostly empirical / forecast-oriented; less normative prescription than Acemoglu or Brynjolfsson
- When normative: more building, less regulation, more openness to disruption — plus the new concrete ask: AI satellite accounts for national statistics
Weak spots / open questions
- "Slow diffusion" + "AGI already here" is an unfalsifiable pair in the short run — per-branch checks now attached (see predictions) so each half can be graded separately
- O-ring framing gained its first real payroll evidence in 2026, but causality (experience premium vs simple hiring freeze) is unresolved
- X-risk engagement remains dismissive rather than steel-manned
- Science-acceleration optimism now has milestones (math ✓, bio trials pending) but he still doesn't specify mechanisms
- His sampling problem: everyone he talks to is in the top percentile of agency
Rich's take
- (your synthesis here)
Delta log
2026-08-28 — v1→v2 migration + validation (batch run)
- Grades: 6 Held / 2 Drifted / 0 Wrong.
- Held Slow-absorption GPT, o-ring (first payroll evidence via Canaries, Aug 2026), science-vs-bureaucracy asymmetry, human-capital premium, geography, healthcare/education laggards.
- Drifted "Growth visible ~2030, not before" — productivity acceleration (2.5%/yr annualized, KC Fed 2026-02-11) and the AI production boom (MR "AI in gdp," 2026-05-28) showed up early; he's amplifying the early numbers himself.
- Drifted The diffusion thesis now coexists with "we're at AGI on most things real humans care about" (DeepMind discussion, July 2025; o3-AGI call Apr 2025) — position restructured: capability closed, diffusion open. Per-branch checks added per learnings rule (tooze 2026-08-24).
- New positions logged: status-redistribution instability; AI satellite accounts; benchmark bias.
- New behavior: Rise and Decline + generative book (Mar 2026); Free Press column; DeepMind/Dwarkesh circuit.
- Most surprising delta: the field's premier diffusion-skeptic became one of the first mainstream economists to declare AGI already achieved — and holds both positions at once.
- Tier: semiannual (positions restructure over quarters, not weeks; daily output but slow thesis drift). Wake triggers: new AI book/essay, MR position shift, GDP/productivity inflection, satellite-account adoption, job change.
Sources
- AI Policy Perspectives: discussion at Google DeepMind (July 2025) — AGI-now, rent/food/healthcare/education frictions, status redistribution
- MR: "AI in gdp" (2026-05-28) — $250B nominal, ~2,600%/yr quality-adjusted, satellite-accounts ask
- The Marginal Revolution: Rise and Decline, and the Pending AI Revolution (2026-03) + generative book
- Dwarkesh Patel podcast #4: "The #1 bottleneck to AI progress is humans" (Jan 2025) · Marcus response to the o3-AGI call (Apr 2025)
- The Free Press: "How to Prepare for the AI Takeover"; MR: economists on AI growth/employment (2026-04)
- Books: The Great Stagnation (2011); Average Is Over (2013); The Complacent Class (2017); Talent (2022); GOAT (2023)
migrated v1→v2 2026-08-28 (weekly validation batch) · backup: _archives/tyler-cowen.html.bak-20260828 · AI & Society domain