Today on The Anvil: the US-Iran ceasefire is cracking under ballistic missile exchanges and threats against allies, while the AI development stack quietly crosses a threshold — agents that learn between sessions, design tools that edit autonomously, and an emerging consensus that the model matters less than the plumbing feeding it.
The US-Iran ceasefire deteriorated sharply on May 27–28. US forces struck an Iranian ground control station in Bandar Abbas and downed four attack drones near the Strait of Hormuz; Iran retaliated by firing a ballistic missile at a US base in Kuwait (intercepted with no damage). Trump then threatened to 'blow up' Oman — a longtime US ally and neutral mediator — if it enters a shipping co-management agreement with Iran over the Strait. The Treasury simultaneously sanctioned Iran's newly created Persian Gulf Strait Authority, which has been attempting to toll commercial shipping. ISW satellite imagery confirms Iran is reconstituting military infrastructure at the Yazd Missile Base since the April ceasefire. Trump told PBS that Iran will receive no sanctions relief in exchange for surrendering enriched uranium, directly contradicting Tehran's negotiating position. Strait traffic remains at 23 ships daily versus the pre-war 125–140.
Why it matters
Three months into this war, the ceasefire is producing more military exchanges than the combat phase. The structural impasse is now clear: the US demands uranium surrender before any sanctions relief; Iran demands economic concessions first. Neither has moved. Trump's threat against Oman — historically the back-channel between Washington and Tehran — destroys a critical diplomatic pathway and signals maximum pressure is replacing negotiation. Iran's creation of a formal Strait toll authority, combined with its military reconstitution during the ceasefire, suggests Tehran is building permanent leverage rather than preparing for compromise. The 80% reduction in Strait traffic continues to pressure global energy markets.
At Code with Claude 2026, Anthropic announced three capabilities for Claude Managed Agents: 'dreaming' — between-session learning where agents identify patterns from prior executions and update their own behavior without human intervention; 'outcomes' — autonomous iteration toward quality rubrics; and multi-agent orchestration coordinating up to 20 parallel specialist agents. These features enable agents that compound in capability over time, accumulating institutional knowledge across teams.
Why it matters
This is the most architecturally significant AI announcement this week. Dreaming closes the loop between production experience and agent improvement — agents that get better from real work rather than requiring offline retraining. For anyone building agent-driven product workflows, this changes the economics: the marginal cost of agent improvement drops toward zero once the infrastructure is in place. The 20-agent orchestration ceiling also sets a practical upper bound for how complex a task graph Claude can manage today. Watch whether competing platforms (Codex, Cursor) match this within weeks.
Figma is materializing its thesis that the traditional design handoff is obsolete, moving the native Design Agent it announced on May 20 into closed beta. Operating directly on the canvas without plugins, the agent can batch-generate layouts, remix designs, and apply feedback across multiple frames in parallel while maintaining full awareness of the file's design system and component library.
Why it matters
You've been following Figma's shift toward an AI-mediated bidirectional workflow. This closed beta puts that strategy into practice: by executing batch operations natively within the canvas, Figma leverages its ultimate moat—deep, file-level context of team comments and component variants—against external AI tools that rely on generic UI generation.
Aikido Security's CTO describes a critical accountability gap as AI coding agents like Claude Code, Copilot, and Cursor autonomously install packages without clear ownership of security responsibility. The company launched Aikido Endpoint to monitor and block malicious packages before installation, addressing the gap between agent autonomy and supply chain trust. This follows last week's TrapDoor campaign and the SymJack exploit, both of which weaponized the AI-agent-to-dependency chain.
Why it matters
The pattern is clear: AI coding agents are making dependency decisions humans never review, and attackers are already exploiting this. Aikido Endpoint represents the first purpose-built security layer sitting between agent action and package installation — treating the agent itself as an attack surface rather than a trusted actor. For any team shipping code through AI agents, the question is no longer whether to add security monitoring but which layer to intercept at. This is becoming load-bearing infrastructure.
CMA CGM, the world's third-largest container shipping line, is rolling out MAIA — an agentic AI platform co-developed with Mistral AI — to all 80,000 employees starting June 1. The platform orchestrates AI agents connected to internal business systems for ETA prediction, demand forecasting, vessel routing optimization, and warehouse labor allocation. CMA CGM has already deployed 55+ AI projects and identified 200+ use cases across its shipping, logistics, and media divisions.
Why it matters
This is agentic AI at genuine enterprise scale in logistics — not a pilot, not a press release about future plans, but a full workforce rollout with measurable operational scope. The Mistral AI partnership is notable: a major logistics operator chose an open-weight European model over US frontier providers, suggesting that for operational AI in regulated industries, model controllability and data sovereignty may outweigh benchmark performance. The 200+ identified use cases across routing, forecasting, and labor management represent the broadest single-company deployment of supply chain AI agents documented this year.
Figure AI's humanoid robots completed a 17-hour package sorting test processing 22,000 packages at ~1,300/hour without interruption. Separately, the company signed a commercial deployment agreement with Catalyst Brands (JCPenney, Aéropostale, Brooks Brothers) for their Reno distribution center. Figure has scaled production to one robot per hour with 350+ third-generation units deployed.
Why it matters
These are the most concrete, quantified metrics from a humanoid robot in sustained warehouse labor. The combination of a verified endurance test and a signed commercial contract — not a pilot, not an MOU — moves Figure AI from demonstration to procurement conversation. At 1,300 packages/hour sustained over 17 hours, the unit economics start to make sense for warehouse operators, especially in labor-constrained markets. The Catalyst Brands deal also shows humanoid robots entering mid-market retail logistics, not just Amazon-scale operations.
Stratasys announced a definitive agreement to acquire Markforged from Nano Dimension for $42.5 million in cash, expected to close in H2 2026. The deal brings Markforged's continuous carbon fiber printing technology and Digital Forge software ecosystem under Stratasys, targeting aerospace, defense, automotive, and industrial production applications.
Why it matters
Markforged pioneered continuous fiber reinforcement in desktop 3D printing — a capability that's become essential for functional prototyping and end-use parts in defense and aerospace. At $42.5M, this is a fire-sale price reflecting the broader challenges in additive manufacturing commercialization, but it concentrates high-performance composite printing capabilities under a single vendor. For product designers using additive manufacturing, the consolidation simplifies the vendor landscape but raises questions about platform lock-in and continued development investment in Markforged's ecosystem.
A chemical implosion at a paper mill in Washington state killed two people with nine others still unaccounted for as of May 28. Hazardous conditions at the damaged tank are slowing cleanup and recovery efforts.
Why it matters
A major industrial disaster in the state raises immediate questions about workplace safety protocols and chemical handling at manufacturing facilities — echoing the Orange County GKN crisis that just resolved. Emergency response capabilities and regulatory oversight of industrial operations are under renewed scrutiny across the Pacific Northwest.
The Kootenai County Republican Central Committee voted to transfer approximately $64,000 — half its bank account — to the Idaho state party, citing plans to defeat a ballot initiative on abortion rights. The decision came days after moderates gained precinct seats and toppled long-time Chairman Brent Regan, raising questions about whether outgoing leadership is stripping resources from incoming moderates.
Why it matters
This is a significant power shift in one of Idaho's most politically active counties. The financial transfer — executed between a leadership change and formal handover — is a tactic that uses party resources as leverage to shape policy outcomes regardless of democratic results within the party. Whether courts or party rules constrain this kind of move will set precedent for how Idaho county parties manage internal transitions.
The Idaho Transportation Department is implementing multiple construction phases along the I-90 corridor near Coeur d'Alene, including an overnight bridge girder installation at Huetter Road tonight and planned ramp closures on June 1 and June 8. The widening project is expected to continue through 2029.
Why it matters
Practical commute and logistics impact for anyone traveling the I-90 corridor between Spokane and Coeur d'Alene. The multi-year widening project reflects the region's sustained population growth and the infrastructure investment required to support it.
With the GKN Aerospace tank stabilized and the 50,000 evacuated residents back home, the focus in Garden Grove has shifted to accountability. DA Todd Spitzer has launched a criminal probe into GKN's maintenance practices, while a Voice of OC investigation revealed significant understaffing at Cal/OSHA and regional air quality regulators. County officials are exploring reimbursement processes for the evacuees, and Newport Beach and Costa Mesa are dealing with residual trash collection delays following emergency evacuations at regional CR&R facilities.
Why it matters
You're familiar with GKN's history of uncorrected violations and the ensuing SB 954 debate. The new revelation is the scale of the regulatory shortfall: extreme understaffing at Cal/OSHA and SCAQMD means similar industrial hazards across Orange County are likely operating under minimal oversight. Spitzer's criminal investigation marks the first step in translating this specific emergency into a broader regional reckoning over industrial safety.
Ukraine's 'People's Satellite' — a crowdfunded synthetic aperture radar capability operated through a contract with Finnish company ICEYE — has now delivered over 5,900 radar images to military intelligence since September 2022. The $16–17M program, funded by redirecting a 2022 drone campaign, provides all-weather geospatial intelligence at 0.25m resolution that has been credited with supporting strikes on Russian naval assets. Poland and Germany are now adopting similar crowdfunded defense satellite models.
Why it matters
This is a landmark case study in democratized geospatial intelligence. A crowdfunded program achieved operational impact comparable to state-owned satellite systems, demonstrating that commercial SAR constellations can be leveraged for military-grade OSINT at a fraction of traditional procurement costs. The adoption by NATO members suggests this model — public funding of commercial space assets for defense — is becoming a procurement paradigm rather than a wartime improvisation.
Context Engineering Replaces Model Selection as the Core Skill Across Anthropic's agentic coding report, Atlassian's workflow research, and multiple builder analyses, the consensus is hardening: competitive advantage no longer comes from choosing the best LLM but from building the infrastructure — context engines, memory systems, design system registries — that feeds agents the right signal. The model is becoming commodity; the plumbing is the moat.
Agents Graduate from Stateless to Learning Systems Anthropic's managed agents with 'dreaming' (between-session learning), CoreWeave's training-to-inference feedback loop, and Modiqo's deterministic workflow capture all point toward the same shift: agents that accumulate institutional knowledge and improve without human re-training. This is the transition from tool to teammate.
Ceasefire Infrastructure Is Failing Across Multiple Theaters The US-Iran conflict is producing daily military exchanges despite a nominal ceasefire, with ISW satellite imagery confirming Iranian military reconstitution. Control of the Strait of Hormuz — through which 20% of global oil flows — has become the central negotiating lever, with Trump now threatening allies to prevent Iranian co-management.
Supply Chain AI Shifts from Insight to Execution Epicor's agentic ERP agents, CMA CGM's 80,000-employee MAIA platform, and Figure AI's 17-hour warehouse demonstrations all show AI moving past dashboard analytics into autonomous task execution in logistics. The gap between pilot and production is closing — but only for organizations that invested in data architecture first.
Design-to-Code Friction Is Collapsing Figma's native design agent, Claude Cowork for design system maintenance, and the shadcncraft Figma-to-React pipeline represent a convergence: design and code are no longer separate workflows mediated by handoff documents. They're becoming a single loop where AI maintains consistency across both surfaces.
What to Expect
2026-06-01—CMA CGM begins company-wide rollout of MAIA agentic AI platform to 80,000 employees across shipping, logistics, and media divisions.
2026-06-01—I-90 ramp closures begin near Coeur d'Alene for ITD widening project; additional closures scheduled June 8.
2026-06-12—Cursor IPO reportedly scheduled; SpaceX acquisition for $60B still unconfirmed.
2026-06-15—Grok V9-Medium (1.5T parameters) expected public release, trained on Cursor developer workflow data.
2026-08-02—EU AI Act Article 50 transparency obligations take effect — first live compliance deadline despite Omnibus deferral of high-risk system rules to 2027.
How We Built This Briefing
Every story, researched.
Every story verified across multiple sources before publication.
🔍
Scanned
Across multiple search engines and news databases
888
📖
Read in full
Every article opened, read, and evaluated
173
⭐
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