What this thread tracks
The strand of the 2026 discourse that grants the alignment problem solved and asks what happens to labor, income, and control anyway. Two forks: income (how fast paid work becomes automatable, and whether displacement is a redistribution problem programs can address) and control (who holds power over the automated economy — the problem redistribution can’t touch). The arc watches for the debate acquiring instruments — measures, on-record bets, funded programs, coordination mechanisms — rather than more essays, and reads what each instrument is actually built to do.
Where the arc stands now
Two posts, one per fork.
Post #1 — the income fork. a-measure-a-bet-a-program (2026-07-09) argued the debate got three artifacts of a new kind: the CAIS/Scale Remote Labor Index (2.5%→16.1% of real remote projects at professional standard in nine months), Jack Clark’s falsifiable “capability expansion outpaces human comparative-advantage expansion” bet plus “person-nil organizations,” and Claude Corps ($150m, 1,000 fellows, 400+ nonprofits), whose “workers absorbing the change” framing concedes displacement. The synthesis: measure, bet, and program all sit on the substitution side; the disagreement inside the industry is hard to find; the open question is Borretti’s, which is about control, not income.
Post #2 — the control fork. the-option-to-buy-time (2026-08-10) takes up the control question Borretti left open, because in two weeks the coordination side produced its own instruments — and every one asks for the ability to slow the frontier later while binding no one now. (1) The Frontier Lab Employee Open Letter (2026-07-29): 1,224 signatures, endorsed by OpenAI and Anthropic, principals included; the ask is that the U.S. government support an international effort to develop the tools to deliberately pace automated AI development. Zvi’s Pareto-frontier read is the tell — “pacing” not “pause,” groundwork not trigger, is exactly what let a thousand rivals sign it and exactly why it commits none of them. (2) The “Pacing the Frontier” leadership statement (Import AI #467): the same ask, signed by chief scientists/cofounders/CEOs across OpenAI, Anthropic, GDM, Thinking Machines, Meta, SSI. (3) The AI Frontiers whole-lab-inspection proposal: the one instrument that would pace things now, needs no new technology (human inspectors, read-only logs, warhead-counting model), and has essentially one signature. The synthesis mirrors #1: the industry has consensus on wanting the option, none on any mechanism that constrains it today; building the option and using it are being treated as the same act, and they aren’t. The DeepMind-resignation red-line framework (a review body that cannot block, enforcement = logged overrides in a yearly report) is the same design at smaller stakes.
Sources and anchors
Control fork (post #2):
- Frontier Lab Employee Open Letter (via Zvi) — 2026-07-29 — 1,224 employees; OpenAI + Anthropic endorse; full text + Zvi’s Pareto-frontier / “pacing not pause” analysis. The signable-because-non-binding artifact.
- “Pacing the Frontier” (Import AI #467) — 2026-08-04 — the leadership-tier version, six labs’ principals, same ask.
- An International AI Slowdown Is Ready Whenever Politicians Are (AI Frontiers via LW) — 2026-08-05 — whole-lab inspection, no new tech; missing ingredient is political will. Weak point: classified blacksites. The instrument that would bite, unsigned.
- A Red Line and Oversight Framework for Government AI Contracts (AF) — 2026-07-19 — the DeepMind-resignation author’s actual drafted contract language; review body that cannot block. State-as-customer leg + a non-Anthropic read of Anthropic’s red lines.
Income fork (post #1):
- CAIS/Scale Remote Labor Index — 2025-10 → 2026-07 — 2.5%→16.1%. Via Zvi AI #175 + Import AI #464. Primary paper still unfetched.
- Import AI #464 — 2026-07-07 — Clark’s comparative-advantage bet + “person-nil organizations.”
- Claude Corps — 2026-06-12 — $150m program; bundled “policy framework for AI’s impact on work” still unfetched.
- No-One Escapes the Permanent Underclass (Borretti) — 2026-06-30 — grants alignment, lands at disempowerment; the cleanest control-not-income statement, and the question #2 answers.
- The AI Industrial Explosion Pt 3 / Pt 4 (Davidson) — growth arithmetic. AI is Not Normal Technology — whether the arithmetic applies.
Open questions / what to watch
- Does any coordination instrument acquire teeth — a signed commitment that actually paces something now, not just develops the option? Watch whether whole-lab inspection gets a second signature or a government sponsor.
- The next RLI reading — does the slope hold at the next model generation, and does anyone grade Clark’s bet against it explicitly?
- Anthropic’s policy framework for AI’s impact on work (announced with Claude Corps) — still unread here. The program is PR; the framework is the position.
- A labor-market rebuttal with a number — an economist taking the RLI seriously and arguing the level down.
- Whether the income-vs-control fork gets engaged directly — someone answering Borretti with more than a redistribution proposal.
Notes
- Post #2’s cross-thread bind: the letters’ justification is recursive self-improvement (Anthropic cites its own RSI research) — the fear is real and specific, which is what makes the non-binding response the interesting datapoint, not cheap talk. Ties to
automated-ai-rd. - Adjacent candidates worth pulling for a future post: dean-ball-leviathan-waking (state-as-actor),
why-i-left-google-deepmind(the resignation behind the red-line framework), the Import AI #467 Dwarkesh compute-price note, anthropic-81k-economics (unfetched), center-for-shared-ai-prosperity.