hornof · hornof · Aug 18, 2026 · Aug 18, 2026
diff --git a/companies/amazon.md b/companies/amazon.md
@@ -19,7 +19,8 @@ last_updated: 2026-08-12
 
 ## Traction Signals
 
-- *(Stub — created to close a ghost link. Amazon recurs across the wiki via AWS, the Anthropic partnership, and the 4-vendor compute cluster; expand with concrete signals — AWS AI-infra announcements, Anthropic-deal specifics, Bedrock/Trainium — as they're ingested.)*
+- **2026-08-17 — "destroying rare books to train AI"** (404 Media investigation, via [[simon-willison]], [[dailybrief-roundup-2026-08-17]]): reporters tracked a shipment of **scarce/rare texts** that ended at an **Amazon AI-training facility** — physical books consumed (destructively scanned) as training data. A **training-data-provenance + IP/copyright + cultural-heritage** signal: Amazon's buying power is reshaping *what data exists* for training. Brief's read: *"the corpus your model trains on is partly determined by whoever can afford to buy it first."* First concrete non-AWS AI signal on this page; a create-candidate anchor for a future `concepts/training-data-provenance`. *(Investigative report; Amazon response not captured.)*
+- *(Otherwise still a stub. Amazon recurs across the wiki via AWS, the Anthropic partnership, and the 4-vendor compute cluster; expand with concrete signals — AWS AI-infra announcements, Anthropic-deal specifics, Bedrock/Trainium — as they're ingested.)*
 
 ## Resources
 - [[andy-jassy]] — CEO

diff --git a/companies/anthropic.md b/companies/anthropic.md
@@ -134,6 +134,7 @@ Cataloged via [[rubenhassid-anthropic-30-term-map-2026-05]] — single secondary
 - 2026-05-28: **[[claude-opus-4-8|Claude Opus 4.8 launch.]]** Recommended model for [[claude-code|Claude Code]]; Anthropic claims 69.2% on SWE-bench Pro, outperforming GPT-5.5 and Gemini 3.1 Pro on senior-level engineering / writing. Same pricing as Opus 4.7. Bundled with **Dynamic Workflows** preview for parallel subagents — orchestration primitive for multi-agent code migration at scale + self-verifying workflow patterns. Dan Shipper named as Codex→Opus migration claim at launch (marketing-coordinated; track for unsolicited follow-on). Same-day as Series H. — [[claude-opus-4-8-dynamic-workflows-2026-05-28]]
 - 2026-05-28: **Dynamic Workflows primary** — full Anthropic-primary detail on the new orchestration primitive: **tens to hundreds of parallel subagents in a single session**, new **`ultracode`** Claude Code setting, **adversarial-agent verification + convergence-termination**, persistence across interruption. **Bun Zig→Rust port (Jarred Sumner) as launch case study**: 99.8% test suite passing, ~750K LOC Rust, 11 days, hundreds of agents in parallel with 2 reviewers per file (port not yet in production). Distribution: Claude Code CLI + Desktop + VS Code + API + Bedrock + Vertex + Foundry on day one. Default-on for Max/Team/API; default-off for Enterprise. Explicit *"substantially more tokens"* operator-cost warning + first-run confirmation gate. — [[anthropic-dynamic-workflows-primary-2026-05-28]]
 - 2026-05-29: **Run-rate revenue hits $47B** ([[simon-willison|Willison]] surfacing). Prior wiki-tracked anchor was $44B as of May 9 ([[aakashgupta-anthropic-growth-acceleration-2026-05-09]]); **+$3B / +6.8% in three weeks**. Brief insightful framing flags the **run-rate vs ARR distinction**: *"shipping run-rate metrics instead of actual ARR is telling. They're signaling adoption velocity to investors, not profitability."* The $47B is the **underlying-fundamentals component** of the same-week capital + capability + values bundle (Series H + Opus 4.8 + ad-free positioning + Founder's Playbook + Dynamic Workflows primary + $47B run-rate = **6-event same-week strategic-coordination bundle**). Verification-pending: ARR-vs-run-rate methodology; customer concentration in the May acceleration; period covered. — [[anthropic-47b-runrate-willison-2026-05-29]]
+- 2026-08-17: **Annualized revenue surges to $65B** (TechCrunch, [[dailybrief-roundup-2026-08-17]]). Prior wiki-tracked anchor was **$47B run-rate** (2026-05-29) — **+$18B / +38% in ~2.5 months**, the curve still bending up rather than decaying (consistent with the [[aakashgupta-anthropic-growth-acceleration-2026-05-09|$44B-in-17-months]] acceleration thesis). Brief's structural read: *"when inference economics work hard enough to swing $18B ARR in two months, the bet isn't on better weights anymore — it's on who owns the deployment layer"* — the same *value-shifts-to-the-decision-layer* thesis the [[ai-margin-collapse]] thread tracks (and which the [[stripe|Stripe/OpenRouter]] acquisition prices from the infra side the same day). Same run-rate-vs-ARR caveat as the $47B figure; TechCrunch-reported, methodology unstated. — [[dailybrief-roundup-2026-08-17]]
 - 2026-05-29: **MIT CSAIL Alex Zhang's recursive-language-model research connects to Anthropic's Scaling Managed Agents + Dynamic Workflows** (via Digg). **First wiki-captured external-research → Anthropic-agent-systems direct lineage claim**; cross-confirms Dynamic Workflows as research-backed rather than purely product-engineered. Primary fetch pending. — [[dailybrief-roundup-2026-05-29]]
 - 2026-05-29: **Lenny Rachitsky dream-companies survey: Anthropic #1.** Combined-platform (X + LinkedIn) survey — *"Anthropic running away with it right now."* Also: Google over OpenAI; Vercel/Linear/Every/PostHog overperforming; *"so many people want to start their own company."* **4-surface convergence on Anthropic-as-talent-magnet by late May 2026** (alongside [[brianlamanna-paraform-talent-density-2026-05|Paraform talent density #3]] + [[techcrunch-anthropic-ramp-business-customers-2026-05-13|Ramp paid-business-customer #1]] + [[aakashgupta-anthropic-growth-acceleration-2026-05-09|$44B-in-17-months revenue acceleration]]). Dream-company surveys are *lagging* indicators — Anthropic leading now means the underlying-fundamentals lead has been substantial enough for long enough to flip cultural perception. The strategic-coordination bundle now has 7 components (capital + capability + values + curriculum + revenue + workflows + talent-preference). — [[lennysan-dream-companies-survey-2026-05-27]]
 - 2026-05-30: **Series G $340B comparison (brief framing)**: 2026-05-30 brief surfaces *"Anthropic raised at 2.8× their Series G valuation ($340B, Feb 2024) in 15 months."* **Internal inconsistency** — Feb 2024 → May 2026 = 27 months, not 15. Either the date or the months are off; resolve before propagating. If 2.8× over 27 months holds (~2.0× annualized appreciation), pairs with the [[aakashgupta-anthropic-growth-acceleration-2026-05-09|$1B→$44B ARR in 17 months]] revenue acceleration (26× annualized) — **valuation appreciation growing slower than revenue acceleration** → multiples compressing, not expanding. First wiki-captured private-market multiple-compression-signal for Anthropic. Verification-pending: Series G timing + valuation. — [[dailybrief-roundup-2026-05-30]]

diff --git a/companies/stripe.md b/companies/stripe.md
@@ -12,6 +12,7 @@ last_updated: 2026-06-30
 
 ## Traction Signals
 
+- **2026-08-17: Stripe acquires OpenRouter for $7B** ([[dailybrief-roundup-2026-08-17]], Latent Space AINews) — payments infrastructure absorbing the **model-routing / decision layer** (OpenRouter is the multi-model routing abstraction that sits between apps and LLM providers). The clearest signal yet that *"the real margin in AI infra isn't owning GPUs — it's owning the decision layer that gets between the app and the model"* — a direct instantiation of the [[ai-margin-collapse|margin-collapse corollary]] that value shifts to the distribution/integration layer as inference commoditizes. Extends Stripe's AI-agent-payments-substrate positioning from *payments rails* to *model-access rails*. Lands the same day as [[anthropic|Anthropic's $65B annualized revenue]] — both point at the deployment/decision layer as the value edge. **Create-candidate `companies/openrouter`.** *(Acquisition reported via AINews; deal terms/close-date primary not fetched.)*
 - **2026-06-30: STRUCTURALLY MAJOR canonical-Visa-Stripe-Coinbase canonical-OpenUSD-stablecoin canonical-3-vendor-backer canonical-cluster** ([[dailybrief-roundup-2026-06-30]]) — canonical-Open-Standard canonical-stablecoin canonical-share-economics-with-users canonical-thesis; canonical-Tempo canonical-day-one-issuer.
 - **2026-06-24: canonical-Intercept canonical-$500M canonical-pathogen-defense-fund canonical-Stripe-and-Anthropic-backed** ([[dailybrief-roundup-2026-06-24]]) — canonical-Stripe-canonical-LP canonical-AI-applied-bio canonical-positioning.
 - **May 2026: canonical-Cloudflare-Stripe-Projects canonical-AI-agent-payments canonical-co-launch** ([[anysphere|Anysphere context]] verification-pending; canonical-protocol for canonical-agent-driven-account-creation + canonical-domain-purchase + canonical-deployment).

diff --git a/concepts/ai-margin-collapse.md b/concepts/ai-margin-collapse.md
@@ -20,7 +20,9 @@ It's the unit-economics lens for evaluating any AI-applied company you'd join or
 - **Lifecycle-phases counter-framing** ([[techcrunch-open-source-not-hurting-anthropic-2026-07-07|TechCrunch, 2026-07-07]]): open-weight and closed frontier models occupy **different lifecycle phases, not the same competitive lane** — which is why open models haven't dented Anthropic's business *yet*. This is a **timing disagreement, not a refutation**: Alderson says the collapse triggers when a credible open peer arrives; TechCrunch says the peer isn't competing for the same (frontier) work yet, so the collapse is deferred until the phases converge. The load-bearing word in both is *"yet."* Corroborated by [[latent-space-field-guide-to-fable-2026-07-08|AutomationBench-AA]]: best open-weight ([[glm-5-2|GLM-5.2]]) scores 27.8% vs Fable 5's 48.6% — a real capability gap on agentic-automation work, consistent with "different phase."
 - **Compute-price counterforce** ([[dwarkesh-patel|Dwarkesh Patel]], *"Why compute might get 10x+ more expensive,"* 2026-07-29, [[dailybrief-roundup-2026-07-29]]): if software-engineer-grade capability is market-priced to human salary, **H100 spot pricing is ~15× below equilibrium** — implying compute costs *rise*, not fall, as capability approaches labor-substitution. This cuts against naive "inference gets ever-cheaper" margin math: today's inference margins may be an artifact of *underpriced* compute. The open-question is whether the labor-arbitrage demand ceiling is hit **before or after** someone learns to run human-level reasoning on ~2 orders of magnitude less compute.
 - **Reasoning-trace theft as a distillation/cost-bypass vector (2026-08-11/13)** ([[dailybrief-roundup-2026-08-13]], [[simon-willison|Willison]] + Latent Space): researchers show frontier models (Anthropic/OpenAI/Google) leak **encrypted chain-of-thought** that persists across sessions/users, and an attacker can **replay a pricey model's reasoning trace into a cheaper sibling** to inherit the reasoning *without paying for it*. A structural twist on the margin thesis: it's not just *open-weight parity* undercutting price — the **frontier's own reasoning output becomes a cheap input** (a reproducibility feature that accidentally enabled a cost-bypass attack). Sharpens the [[reverse-information-paradox|who-captures-the-learning]] question and the distillation-governance thread ([[frontier-ai-governance]] / [[us-treasury-china-ai-sanctions-threat-2026-07-21|Bessent]]).
-- Track: independent GLM-vs-Opus benchmarks; whether frontier labs cut inference prices in response; open-weights adoption in production; whether compute spot-prices climb toward Dwarkesh's labor-anchored equilibrium.
+- **Value-shifts-to-the-decision-layer, priced (2026-08-17)** ([[dailybrief-roundup-2026-08-17]]): the margin thesis's *"value moves to the integration/product layer"* corollary got a **$7B price tag** — [[stripe|Stripe acquired OpenRouter]] (the multi-model routing abstraction). Brief's read: *"the real margin in AI infra isn't owning GPUs — it's owning the decision layer that gets between the app and the model."* Pairs with [[anthropic|Anthropic's $65B annualized revenue]] the same day (*"the bet isn't on better weights anymore — it's on who owns the deployment layer"*): both the applied-infra M&A and the frontier-lab revenue curve point at **distribution/decision** rather than raw model quality as the durable edge.
+- **Cost-capability retrospective (2026-08-17)**: *"From BERT to Frontier Agents: Eight Years of LM Progress"* (arXiv 2608.13675) reports **~6× annual coding-ability improvement since late 2024** and **GPT-5.6 Luna at flagship performance for $1–6/M tokens** — a concrete anchor for the falling-price side of the curve, alongside [[qwen|Qwen 3.8 27B]] matching GPT-5.6 Luna's AA Index score at laptop scale. *(arXiv retrospective; not independently reproduced.)*
+- Track: independent GLM-vs-Opus benchmarks; whether frontier labs cut inference prices in response; open-weights adoption in production; whether compute spot-prices climb toward Dwarkesh's labor-anchored equilibrium; whether the routing/decision layer ([[stripe|Stripe/OpenRouter]]) captures the margin the model layer loses.
 
 ## Key Papers / Posts
 

diff --git a/concepts/ai-vulnerability-discovery.md b/concepts/ai-vulnerability-discovery.md
@@ -115,6 +115,10 @@ Joins the **4-primitive agent-output integrity cluster**:
 
 **Pattern**: each vendor publishes domain-specific primitives at the agent-security layer; not yet a single canonical framework. Pattern-watch: do Microsoft / Anthropic / Meta ship comparable Lockdown-Mode-equivalent primitives?
 
+## Realized risk — AI-generated code as the supply-chain vector: Copilot Autofix → Snowflake Jira (2026-08-17)
+
+Wiz research (*"red-agent-snowflake-copilot-cicd-bug,"* [[dailybrief-roundup-2026-08-17]]) documents **GitHub Copilot "Autofix" generating code that allowed compromise of Snowflake's Jira** — an AI-generated fix that introduced a **supply-chain security hole in real production CI/CD**. Flips the offense/defense frame on this page: the [[willison-firefox-claude-mythos-2026-05|Mozilla defender-side throughput jump]] shows AI *closing* vulnerabilities at scale, while this shows AI *opening* them when its output is trusted into the build without review — the [[simon-willison|Willison]] *"as agents get more reliable I stop reviewing every line"* discipline-decay risk, realized as a logged incident on a fintech/SaaS target. Concrete instance of the *AI-code-trust* failure mode: the defender surface and the attack surface are the same generated artifact. *(Wiz blog; primary not deeply fetched.)*
+
 ## Related Concepts
 
 - [[verifiability-and-jagged-intelligence]] — vulnerability discovery is a verifiable domain (did the exploit work?), so RL training should drive rapid capability growth here; this concept gives the mechanism for why the offense/defense balance is unstable

diff --git a/log.md b/log.md
@@ -468,3 +468,5 @@ Cross-cutting synthesis: **three independent voices converged on the ITSM-first
 2026-08-14 | ingest | _raw Andrew Ng "AI Engineering Skills Map" | pages touched: concepts/ai-engineering-skills (NEW), sources/andrew-ng-ai-engineering-skills-map-2026-08-14 (new), sources/raw-batch-roundup-2026-08-14-skillsmap (new), people/andrew-ng (skills-map fold + bump 07-10→08-14), index (ai-engineering-skills). SOURCE: 1 _raw drop (no Daily Brief). Andrew Ng "The AI Engineering Skills Map" (981w, X) — data-backed (10k+ job postings + dozens expert/hiring-manager interviews + surveys) taxonomy of 4 core AI-engineering skills: (1) building+deploying AI apps (evals+error-analysis-loops core sub-skill), (2) SWE fundamentals (steer agents in "precise language of software engineering"), (3) using coding agents (context-mgmt/verifiers/multi-agent-orchestration/avoid-prod-DB-disaster/spec-when-worth-it/keep-trying-new-tools), (4) shaping the build ("work shifting toward deciding what should be in the spec" + product-sense + ownership/agency). Framed as SKILLS not the "AI Engineer" role (like cloud skills). OWNER-RELEVANT: the wiki's hands-on-ramp mission as a curriculum. NEW concept ai-engineering-skills (canonical + data-backed + owner-relevant + anchors a cluster). 1:1 convergence w/ wiki spine (loop-engineering/context-engineering/graph-engineering/FDE/eng-leadership) — mainstream-authority ratification. Ng "shaping the build" = same judgment/agency FDE + Garry-Tan-experienced-founder price at premium (3-source convergence on judgment-as-durable-human-layer). Per-skill deep-dives promised (track for fuller map). 0 delta ghosts. iCloud glob caveat: verified via per-target existence checks.
 
 2026-08-16 | ingest | Daily Briefs/2026-08-16.md | pages touched: sources/dailybrief-roundup-2026-08-16 (new), concepts/graph-engineering, companies/xai, people/simon-willison, concepts/frontier-ai-governance. SOURCE: Daily Brief 2026-08-16. MOST HEADLINES RE-SURFACES (dedup, no action): Greenblatt/Dwarkesh RSI (ryan-greenblatt #242/#237), Grok 4.6+Grok@Bot (xai #242), Import AI 468 / PostTrainBench / 23-RSI-ideas (jack-clark #239), Redeploying Fable 5 + jailbreak-severity framework (anthropic-redeploying-fable-5... 2026-06-30), Claude system-prompts docs (routine). NET-NEW FOLDS: (1) Anthropic "Patterns and problems in emerging multi-agent systems" research → graph-engineering (first-party failure-mode taxonomy; "shipping coordination without understanding it"); (2) Grok CSAM real-harm case (TechCrunch 08-15) → xai Community Sentiment (structural safety gap; ship-image-gen-without-gates pattern, pairs w/ grok-build no-consent smell); (3) Willison local-inference cluster (CORS Chat local-LLM playground built in hours w/ GPT-5.6-Sol — "local inference stopped being a researcher thing" + "hallucinate don't classify" tagging + sqlite-utils 4.2) → simon-willison (feeds open-weights/local-models thread); (4) Anthropic text watermark shipped → frontier-ai-governance (provenance lever: SRO proposal→shipped feature). WATCH-ITEMS (noted, not folded): arXiv 2608.13122 AI-assisted GPU porting of 250k-line legacy weather code (owner-relevant legacy-modernization; unclear if generalizes), fifth-grade-text-only training (capability floors), Flue 2 "React for Agents" (Fred Schott; create-candidate tools/flue), DeepMind sign-language (recurring-deferred). No new entity pages; no index change. 0 delta ghosts.
+
+2026-08-17 | ingest | Daily Briefs/2026-08-17.md | pages touched: sources/dailybrief-roundup-2026-08-17 (new), companies/anthropic, companies/amazon, companies/stripe, concepts/ai-margin-collapse, models/qwen, concepts/ai-vulnerability-discovery, people/jack-clark. SOURCE: Daily Brief 2026-08-17. NET-NEW FOLDS: (1) Anthropic annualized revenue $65B (TechCrunch) → anthropic (+$18B/+38% vs $47B run-rate May-29; "bet is on who owns the deployment layer"); (2) Stripe acquires OpenRouter $7B (Latent Space AINews) → stripe + ai-margin-collapse ("margin isn't owning GPUs, it's owning the decision layer"; value-shifts-to-distribution corollary priced; create-candidate companies/openrouter); (3) Amazon destroying rare books to train AI (404 Media/Willison) → amazon (training-data-provenance + IP/copyright + cultural-heritage; first non-AWS AI signal on the stub; create-candidate concepts/training-data-provenance); (4) GitHub Copilot Autofix → Snowflake Jira compromise (Wiz) → ai-vulnerability-discovery (AI-generated code as supply-chain vector; realized AI-code-trust failure); (5) Qwen 3.8 27B AA Index 52 = GPT-5.6 Luna, laptop-runnable, "wildly overthinks" (Willison x2) → qwen (efficiency milestone sharpening open-weight-parity); (6) Import AI 469 Science AI + RSI simulator + Zuck pessimism → jack-clark (light). WATCH-ITEMS: Groq chips→neocloud pivot $350M/$3.5B (create-candidate companies/groq); arXiv 2608.13566 benchmark-overfitting critique (no evals concept yet); arXiv 2608.13675 BERT-to-agents 6x/yr coding + GPT-5.6 Luna $1-6/M (cross-ref ai-margin-collapse); Nvidia $1.5B SoftBank DC; tail-latency fix. RE-SURFACES (dedup): Greenblatt RSI (3rd surface), DeepMind sign-language (now shipped, still deferred). No new entity pages; no index change. 0 delta ghosts.
Original file line number	Diff line number	Diff line change
Expand Up		@@ -468,3 +468,5 @@ Cross-cutting synthesis: **three independent voices converged on the ITSM-first
		2026-08-14 \| ingest \| _raw Andrew Ng "AI Engineering Skills Map" \| pages touched: concepts/ai-engineering-skills (NEW), sources/andrew-ng-ai-engineering-skills-map-2026-08-14 (new), sources/raw-batch-roundup-2026-08-14-skillsmap (new), people/andrew-ng (skills-map fold + bump 07-10→08-14), index (ai-engineering-skills). SOURCE: 1 _raw drop (no Daily Brief). Andrew Ng "The AI Engineering Skills Map" (981w, X) — data-backed (10k+ job postings + dozens expert/hiring-manager interviews + surveys) taxonomy of 4 core AI-engineering skills: (1) building+deploying AI apps (evals+error-analysis-loops core sub-skill), (2) SWE fundamentals (steer agents in "precise language of software engineering"), (3) using coding agents (context-mgmt/verifiers/multi-agent-orchestration/avoid-prod-DB-disaster/spec-when-worth-it/keep-trying-new-tools), (4) shaping the build ("work shifting toward deciding what should be in the spec" + product-sense + ownership/agency). Framed as SKILLS not the "AI Engineer" role (like cloud skills). OWNER-RELEVANT: the wiki's hands-on-ramp mission as a curriculum. NEW concept ai-engineering-skills (canonical + data-backed + owner-relevant + anchors a cluster). 1:1 convergence w/ wiki spine (loop-engineering/context-engineering/graph-engineering/FDE/eng-leadership) — mainstream-authority ratification. Ng "shaping the build" = same judgment/agency FDE + Garry-Tan-experienced-founder price at premium (3-source convergence on judgment-as-durable-human-layer). Per-skill deep-dives promised (track for fuller map). 0 delta ghosts. iCloud glob caveat: verified via per-target existence checks.

		2026-08-16 \| ingest \| Daily Briefs/2026-08-16.md \| pages touched: sources/dailybrief-roundup-2026-08-16 (new), concepts/graph-engineering, companies/xai, people/simon-willison, concepts/frontier-ai-governance. SOURCE: Daily Brief 2026-08-16. MOST HEADLINES RE-SURFACES (dedup, no action): Greenblatt/Dwarkesh RSI (ryan-greenblatt #242/#237), Grok 4.6+Grok@Bot (xai #242), Import AI 468 / PostTrainBench / 23-RSI-ideas (jack-clark #239), Redeploying Fable 5 + jailbreak-severity framework (anthropic-redeploying-fable-5... 2026-06-30), Claude system-prompts docs (routine). NET-NEW FOLDS: (1) Anthropic "Patterns and problems in emerging multi-agent systems" research → graph-engineering (first-party failure-mode taxonomy; "shipping coordination without understanding it"); (2) Grok CSAM real-harm case (TechCrunch 08-15) → xai Community Sentiment (structural safety gap; ship-image-gen-without-gates pattern, pairs w/ grok-build no-consent smell); (3) Willison local-inference cluster (CORS Chat local-LLM playground built in hours w/ GPT-5.6-Sol — "local inference stopped being a researcher thing" + "hallucinate don't classify" tagging + sqlite-utils 4.2) → simon-willison (feeds open-weights/local-models thread); (4) Anthropic text watermark shipped → frontier-ai-governance (provenance lever: SRO proposal→shipped feature). WATCH-ITEMS (noted, not folded): arXiv 2608.13122 AI-assisted GPU porting of 250k-line legacy weather code (owner-relevant legacy-modernization; unclear if generalizes), fifth-grade-text-only training (capability floors), Flue 2 "React for Agents" (Fred Schott; create-candidate tools/flue), DeepMind sign-language (recurring-deferred). No new entity pages; no index change. 0 delta ghosts.

		2026-08-17 \| ingest \| Daily Briefs/2026-08-17.md \| pages touched: sources/dailybrief-roundup-2026-08-17 (new), companies/anthropic, companies/amazon, companies/stripe, concepts/ai-margin-collapse, models/qwen, concepts/ai-vulnerability-discovery, people/jack-clark. SOURCE: Daily Brief 2026-08-17. NET-NEW FOLDS: (1) Anthropic annualized revenue $65B (TechCrunch) → anthropic (+$18B/+38% vs $47B run-rate May-29; "bet is on who owns the deployment layer"); (2) Stripe acquires OpenRouter $7B (Latent Space AINews) → stripe + ai-margin-collapse ("margin isn't owning GPUs, it's owning the decision layer"; value-shifts-to-distribution corollary priced; create-candidate companies/openrouter); (3) Amazon destroying rare books to train AI (404 Media/Willison) → amazon (training-data-provenance + IP/copyright + cultural-heritage; first non-AWS AI signal on the stub; create-candidate concepts/training-data-provenance); (4) GitHub Copilot Autofix → Snowflake Jira compromise (Wiz) → ai-vulnerability-discovery (AI-generated code as supply-chain vector; realized AI-code-trust failure); (5) Qwen 3.8 27B AA Index 52 = GPT-5.6 Luna, laptop-runnable, "wildly overthinks" (Willison x2) → qwen (efficiency milestone sharpening open-weight-parity); (6) Import AI 469 Science AI + RSI simulator + Zuck pessimism → jack-clark (light). WATCH-ITEMS: Groq chips→neocloud pivot $350M/$3.5B (create-candidate companies/groq); arXiv 2608.13566 benchmark-overfitting critique (no evals concept yet); arXiv 2608.13675 BERT-to-agents 6x/yr coding + GPT-5.6 Luna $1-6/M (cross-ref ai-margin-collapse); Nvidia $1.5B SoftBank DC; tail-latency fix. RE-SURFACES (dedup): Greenblatt RSI (3rd surface), DeepMind sign-language (now shipped, still deferred). No new entity pages; no index change. 0 delta ghosts.