Why this is in the vault
Lead story is Anthropic's new interactive economic model of three AI-adoption scenarios (modest/substantial/extreme) through 2030, with a headline finding — GDP growth does not translate to worker wage growth once capital captures the gains — that is directly on-thesis for the L5 north-star bet.
Mapping against Ray Data Co
Anthropic's "extreme" scenario (15% annual growth, 17.9% knowledge-worker unemployment, labor's GDP share dropping from 60% to 45%, capital owners capturing most gains) is the sharpest evidenced version yet of the premise underlying project_l5_north_star_strategic_direction and project_income_seat_gap_400k: RDCO's bets are explicitly downstream of agent capability, and the founder has already rejected the employment-seat framing in favor of demand generation as the real constraint. This tool gives a concrete, citable range (modest/substantial/extreme, with ~10,000-respondent public calibration showing most people expect "substantial" not "extreme") to stress-test that bet against rather than relying on vibes — worth a founder pass to see where his own assumptions land relative to the surveyed distribution, and whether the "extreme" capital-capture case changes anything about pacing the phData cert escalator path or the acquisition-funnel thesis.
Secondary: OpenAI's GPT-Image-2.5 (Flare/Sunburst, 50% lower latency, "understands what not to touch" surgical editing) is a incremental image-gen product update — touches MAC/design-asset tooling only at the margin (faster iteration on branded asset edits), no action implied.
⚠️ Sponsorship
Three placements this issue, all recurring members of the previously-confirmed 16+ sponsor rotating pool (no new entrants):
- Finest — standalone "Presented by" block on the Anthropic lead story ("Save up to 70% on AI spend with two lines of code" — cost-routing product, evidence-gated model selection for API and Claude Code).
- Teleport — standalone "Presented by" block on the OpenAI image story (agent-activity observability/audit product, unrelated to image-gen).
- Google Cloud — native ad inside the Signals numbered list (#2: General Motors compute-time/simulation gains on Google Cloud G4 VMs with NVIDIA RTX Pro 6000).
The masthead "In Partnership with" slot is present again but did not resolve to a legible name this issue — it renders as <img alt="partner_image"> with no identifiable filename or link text, reverting to the unresolved state seen 2026-09-03 through 09-08 (the 09-09 issue's QA.tech resolution was evidently a one-off, not a stable format change). None of the three identified sponsors touch the lead story's actual economic-modeling claims — adjacent-market placements (cost/observability tooling), consistent with the pattern across prior issues.
Curation section
Top News
- Anthropic models three 2030 scenarios for AI's effect on US jobs and wages — modest (GDP +1.6%, wages stable), substantial (AI handles ~half of knowledge work, GDP +8.3%, wages flatline, career switching required), extreme (15% annual growth, doubling the economy every 4.5 years, 17.9% knowledge-worker unemployment, labor share of GDP 60%→45%). Public survey of 10,000+ Americans shows most expect "substantial"; only ~10% expect "extreme." This is the lead/mapped item above.
- OpenAI ships GPT-Image-2.5 (Flare + Sunburst) — 50% lower latency than GPT-Image-2, targeted "surgical" editing (change one element, preserve the rest) for ecommerce/ad/design use cases. Noted above, weak mapping.
Signals
- Sony AI open-sourced a tool that predicts undiscovered scientific facts.
- General Motors cut compute time 38%, boosted AV simulation throughput 40% on Google Cloud (sponsored placement, see above).
- Amazon's internal tool predicts A/B test winners with 75-90% accuracy before real users see the test — same "simulate users before shipping" pattern as the newsletter's own framing device this issue.
- A new paper derives a theoretical speed limit on how fast neural networks learn from data.
- A developer built a Face ID clone for Mac using only a webcam.
- Cognition raised $2B at a $48B valuation as Devin's revenue nears $900M.
- Reasoning models produce fractal patterns when solving hard problems, offered as a partial explanation for "overthinking" and 10x token-cost variance between near-identical prompts (Sudoku/mazes/math/ARC-AGI benchmarks, code on GitHub under GilpinLab/loopscape).
No deep-fetches this issue: the plaintext body strips outbound hyperlinks (only "READ MORE" anchor text survives with no resolvable URL), and none of the curated items carried a specific-enough hook plus a resolvable third-party domain to clear the 2-link-max curation bar.
Related
- [[2025-12-04-moonshots-ep212-agi-timeline-job-automation]]
- [[2026-03-07-moonshots-ep236-andrew-yang-ubi]]
- [[project_l5_north_star_strategic_direction]]
- [[project_income_seat_gap_400k]]