Today on The Anvil: the US naval blockade of Iranian ports goes live as oil prices spike toward $150/barrel, AI coding tools converge into composable stacks rather than competing monoliths, and Anthropic overtakes OpenAI in valuation for the first time. Twelve stories across geopolitics, AI infrastructure, local news, and supply chain intelligence.
Building on yesterday's talk collapse: CENTCOM implemented a full naval blockade at 14:00 GMT April 13. New developments today — oil surged to ~$150/barrel; the UK explicitly refused to participate, instead convening a 40-nation freedom-of-navigation coalition; 230 tankers are now trapped in the strait; and Iran's IRGC warned of a 'severe' response. An Asia Times legal analysis finds both Iran's toll-booth regime and the US blockade violate UNCLOS, but neither side will go to the ICJ. The blockade removes 1.7–1.84M barrels/day of Iranian crude, compounding existing supply disruptions.
Why it matters
The UK's refusal is the most significant new diplomatic signal — it narrows the coalition supporting the blockade and raises sustainability questions. The Soufan Center's finding that sanctions relief and frozen assets were actually negotiable before talks collapsed is also new: this wasn't a purely irreconcilable impasse, which makes the escalation choice more consequential. First vessel interception and Iran's military response remain the critical 48-hour watch items.
Anthropic's ARR surge from $9B to $30B (flagged in last week's Treasury/Fed briefing coverage) has now translated to a secondary market valuation overtaking OpenAI — $863.6B vs. $846.1B. New details: institutional investors are reportedly refusing OpenAI shares and queuing for Anthropic stock; VCs at HumanX unanimously favored Anthropic. The structural explanation: OpenAI's 98% free ChatGPT user base generates compute costs without proportional revenue, while Anthropic's B2B focus yields higher margins.
Why it matters
The ARR figure was already in memory; the valuation crossover and institutional capital preference are new. Secondary market dynamics are a leading indicator — when money flows this directionally, it signals conviction about platform economics, not just hype. The competitive reversal also recontextualizes Anthropic's crawling ethics controversy (Story 10): a company this dominant faces proportionally higher scrutiny.
MiniMax released M2.7, an open-source MoE model scoring 56.22% on SWE-Pro — the harder production benchmark tracked here (vs. the inflated SWE-bench Verified scores) — matching GPT-5.3-Codex. The model participated autonomously in 100+ iterations of its own optimization, achieving 30% gains through self-improvement loops, and demonstrated 3-minute production incident debugging. The broader open-source frontier (GLM-5.1, Gemma 4, Mistral Small 4) now achieves single-digit gaps vs. proprietary leaders at 1/50th the price.
Why it matters
The SWE-Pro score matters precisely because previous coverage established the 93.9% vs. 23% benchmark gap — M2.7's 56.22% on the harder test is a meaningful data point, not a marketing claim. More significant is the self-improvement architecture: 100+ autonomous optimization iterations producing 30% gains is the first concrete evidence of productive AI self-optimization at this scale. Cost/performance parity with proprietary models makes vendor lock-in increasingly hard to justify.
Cloudflare data reveals Anthropic has a crawl-to-refer ratio of 8,800 to 1 — the highest among major AI companies — meaning its bots scrape 8,800 pages for every single referral sent back to publishers. The report highlights how generative AI fundamentally breaks the historic internet bargain where search engines sent traffic in exchange for crawling rights.
Why it matters
This is a sharp counterpoint to Anthropic's 'responsible AI' positioning, arriving the same week it overtook OpenAI in valuation. The data challenges the implicit social contract between AI companies and content creators: if AI companies extract web content at massive scale with near-zero traffic return, the economic incentives that sustain content creation collapse. For builders relying on AI-generated outputs, this signals potential friction — publishers are already pushing back with robots.txt blocks, and regulatory responses are likely. The tension between AI companies needing training data and publishers needing traffic has no current resolution.
After weeks of tracking Cursor 3's agent-centric redesign and the benchmark-vs-production gap, the market structure is now clarifying: rather than a winner-take-all outcome, a composable three-layer stack is emerging — orchestration (Cursor), execution (Claude Code, Codex), and review (cross-provider). OpenAI published an official Codex plugin for Claude Code to formalize this layering. New market share figures: Anthropic now commands 42–54% of enterprise code generation vs. OpenAI's 21%; 98% of professional developers report regular AI tool usage; 41% of all code is now AI-generated.
Why it matters
The architectural implication is new: tool selection is no longer binary but structural. Teams can swap execution layers without rebuilding orchestration, reducing vendor lock-in risk. The 41% AI-generated code figure makes this infrastructure-critical. Composability also resolves the tension between Cursor's benchmark lead and Claude Code's enterprise share — they're complements, not competitors.
Cursor's technical report on Composer 2 (the Kimi K2.5-based model you've seen benchmarked) reveals the training infrastructure behind it: asynchronous RL on live production feedback completing full cycles in 5 hours. Key mechanisms: self-summarization for long-horizon context, nonlinear reward shaping from emoji reactions and follow-up edits, MoE router replay. Measured outcomes: 2.28% higher edit persistence, 3.13% fewer dissatisfied follow-ups versus baseline.
Why it matters
This explains why Cursor's benchmark advantage translates to production — the model is trained directly on production retention signals, not synthetic tasks. The 5-hour feedback loop is a structural competitive moat: it improves faster than any competitor on traditional training cycles. For teams building agentic systems, the self-summarization and harness-aligned training patterns are immediately applicable engineering insights.
Google released Vibe Coding XR, a Gemini-powered tool that generates functional VR/AR prototypes from text prompts in under a minute using modular XR Blocks. Natural language orchestrates physics, interactions, and UI without traditional scripting. Currently available on Android XR hardware (Samsung Galaxy XR). Separately, Gemini also added interactive 3D model generation directly within the chat interface — users can request and manipulate molecular models, orbital simulations, and other 3D visualizations without leaving the conversation.
Why it matters
Google is extending the 'vibe coding' paradigm into spatial computing and 3D visualization — domains where prototyping friction has been exceptionally high. For product designers working at the physical-digital intersection, text-to-XR prototype in under a minute collapses a multi-day workflow into an iteration cycle. The in-chat 3D generation is arguably more immediately useful: it means Gemini can now serve as a visual communication tool for spatial concepts without switching to CAD or visualization software. Both features are positioned as exploration tools, not production replacements, but they lower the barrier to spatial ideation significantly.
PwC's 1,217-company study adds the positive case to last week's BCG finding (only 7% tangible returns): the 20% who succeed capture 74% of returns and deliver 7.2x greater AI-driven financial performance. Top performers build data foundations before deploying agents — matching BCG's sequencing recommendation. John Deere's See & Spray (59% herbicide savings, 1M+ acres) is the flagship production-grade example. Six foundational capabilities identified: strategy, data, technology, workforce, governance, innovation.
Why it matters
The BCG and PwC studies now form a coherent picture: most fail (7% tangible returns), but those who sequence correctly win big (7.2x). The six-capability framework serves as a diagnostic against BCG's finding that fragmented data (78% cite forecast inaccuracy) is the root cause of failure. The John Deere case is the kind of concrete outcome that separates real supply chain AI value from pilot purgatory.
With freight costs already up 16–17% YoY and diesel above $5.40/gallon regionally, GasBuddy analyst Patrick De Haan puts 85% of current Spokane pump-price increases on the Hormuz disruption, with $5–7/gallon projected depending on strait status. Washington's Cap-and-Invest program adds ~40 cents/gallon on top.
Why it matters
The 85% Hormuz attribution is the new and locally actionable figure — it clarifies that state policy (40 cents) is noise relative to the geopolitical signal. For Inland Northwest farmers, freight operators, and service workers, this connects directly to the blockade's household-level consequences.
Spokane City Council is considering regulatory reforms that would eliminate the $60 annual permit for short-term food truck operators, remove the paved-parking-lot requirement, and — most significantly — scrap the requirement that food trucks obtain written permission from adjacent property owners before operating downtown. The Downtown Spokane Partnership and Spokane Regional Health District have raised concerns about enforcement and unintended consequences.
Why it matters
This is a meaningful urban activation debate. The adjacent-property-owner permission requirement effectively gives existing businesses veto power over mobile competitors — removing it could significantly increase downtown foot traffic and street-level activity. The tension between reducing regulatory barriers for small operators and maintaining health/safety oversight reflects a broader national trend in cities rethinking how they regulate public space. Worth watching for how it shapes downtown Spokane's pedestrian experience.
MizarVision, a Chinese geospatial firm, correlated publicly available ADS-B signals from KC-135 and KC-46 tankers with satellite imagery to reconstruct specific US strike operations against Iran during Operation Epic Fury — demonstrating real-time monitoring of classified military operations from open-source data.
Why it matters
The methodology (ADS-B correlation + temporal satellite analysis) is the same pattern used by the analysts who identified the downed F-15E last week — but now applied by a state-linked Chinese firm to live US force projection. This is the first case demonstrating China's geospatial intelligence capabilities against active US operations, and it arrives against the backdrop of Planet Labs' 14-day Middle East imagery blackout, which suggests the US anticipated exactly this kind of exposure. The USS Michael Murphy's 43-day AIS silence also looks like a direct response to this threat vector.
Newly revealed FISA Court opinions show the FBI continued to systematically circumvent Section 702 compliance through 2025 using undisclosed 'filtering tools' to conduct US person queries without tracking or approval — FBI, NSA, and CIA all used similar tools, leaving 2025 query counts incomplete. Section 702 reauthorization now faces bipartisan congressional resistance, with a classified court ruling raising overreach concerns.
Why it matters
The pattern — agencies finding technical workarounds that compliance tools fail to detect — is structurally consistent with the OSINT infrastructure bifurcation tracked here: surveillance capabilities outpace governance on both sides. The bipartisan resistance creates a genuine legislative window for reform, possibly the first meaningful check on bulk surveillance since Snowden. The timing, concurrent with China publicly demonstrating ADS-B-based military tracking, adds irony: US operational security is being penetrated via open-source methods while domestic surveillance oversight fails internally.
Blockade Economics Cascade From Hormuz to the Inland Northwest The US naval blockade of Iranian ports is no longer abstract geopolitics — oil hit $150/barrel, Spokane gas analysts project $5-7/gallon, and freight costs (already up 16-17% YoY) face another supply shock. The supply chain, energy, and local news threads are now one story.
AI Coding Tools Are Composing, Not Consolidating Rather than a single winner emerging, Cursor, Claude Code, and Codex are stratifying into orchestration, execution, and review layers. Teams are building multi-tool stacks optimized by task type — a modular architecture replacing the monolithic IDE paradigm.
Open-Source AI Models Reach Functional Parity With Proprietary Frontier MiniMax M2.7, GLM-5.1, Gemma 4, and Mistral Small 4 now deliver 90%+ of frontier performance at a fraction of the cost. The model selection question has shifted from 'which is best' to 'which is best for my budget, latency, and data sensitivity constraints.'
The 7% Problem: AI Adoption Stalls Between Pilot and Production Multiple data points converge — BCG's 7% tangible returns, PwC's 20% capturing 74% of value, Bain's satisfaction gap between scaled and piloting orgs. The pattern is clear: AI investment is surging but production deployment remains the bottleneck.
Surveillance Infrastructure Under Simultaneous Expansion and Scrutiny FBI FISA violations continue undetected, Section 702 renewal faces congressional obstacles, and a Chinese OSINT firm publicly tracked US bomber movements from open ADS-B data. The tools of surveillance and counter-surveillance are both proliferating faster than governance frameworks can adapt.
What to Expect
2026-04-14—US naval blockade of Iranian ports fully operational — watch for first interception incidents and oil market response at Monday open
2026-04-16—Newport Beach International Boat Show opens at Lido Marina Village (runs April 16-19); Corona del Mar 3rd Thursdays inaugural event
2026-04-21—Approximate expiration of the two-week US-Iran ceasefire — key inflection point for renewed hostilities or extension
2026-04-29—BLM North Idaho recreation site closures continue through late May — Huckleberry Campground and multiple boat launches affected
2026-08-12—Coeur d'Alene city council deadline for Kootenai County Multi-Jurisdictional All-Hazard Mitigation Plan adoption
How We Built This Briefing
Every story, researched.
Every story verified across multiple sources before publication.
🔍
Scanned
Across multiple search engines and news databases
535
📖
Read in full
Every article opened, read, and evaluated
133
⭐
Published today
Ranked by importance and verified across sources
12
— The Anvil
🎙 Listen as a podcast
Subscribe in your favorite podcast app to get each new briefing delivered automatically as audio.
Apple Podcasts
Library tab → ••• menu → Follow a Show by URL → paste