The Manifest
The US-Iran Islamabad MOU signals a potential Strait of Hormuz reopening with direct shipping cost implications, while the inference compute market heats up with Baseten's $1.5B raise and AI talent reshuffles as Noam Shazeer crosses from Google to OpenAI.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01How the 14-Point U.S.-Iran MOU Could Reshape Global Supply ChainsFreightWaves
- 02Google's Gemini co-lead Noam Shazeer joins OpenAI after two-year return stintThe Decoder
- 03AI inference startup Baseten reportedly raising $1.5B months after its last mega-roundTechCrunch AI
- 04Imports flow into Port of Los Angeles with 'window of stability' openSupply Chain Dive
- 05Flexport: New tariff wave could replace expiring trade duties by late JulyFreightWaves
Models & Releases
5 storiesPerplexity Launches Brain, a Self-Improving Memory System That Builds a Context Graph of an Agent's Work and Learns Overnight
An agent that reviews its own session history overnight and adjusts based on what succeeded or failed is a meaningful step toward AI that compounds institutional knowledge without retraining. Watch whether this pattern migrates into workflow automation tools where process context accumulates over months.
Anthropic brings Artifacts to Claude Code, letting teams share live pages from coding sessions
Session outputs that auto-update and carry version history close the gap between AI coding work and team documentation. For ops and tech teams deploying internal AI tools, this reduces the cost of auditing what an AI agent actually produced.
Adobe adds AI agents to Photoshop, Premiere, and more Creative Cloud apps
Adobe wiring an agent layer across its core Creative Cloud apps means multi-step tasks described in plain language rather than navigated via menus. It is an early signal for how enterprise software suites are shifting from feature sets to task-automation layers more broadly.
ChatGPT's new health upgrade beats doctor-written answers, OpenAI says
A 71% drop in health-statement error rates through domain-specific tuning illustrates how fast specialization is narrowing the gap between AI and credentialed professionals. The same fine-tuning pattern is already moving into compliance review, contract analysis, and procurement.
Vercel Releases Eve: An Open-Source AI Agent Framework Where Each Agent is a Directory of Files Mapped to Capabilities
An agent framework built on plain file directories with durable execution, approvals, and evals built in is a practical architecture for teams that need internal agents to be auditable. Worth evaluating alongside other orchestration options if you are building workflow automation on open infrastructure.
Supply Chain & Ops
5 storiesHow the 14-Point U.S.-Iran MOU Could Reshape Global Supply Chains
If the deal holds, a Hormuz reopening removes the routing detour that added days and cost to Middle East energy and cargo flows. Planners should model both scenarios now, since 60 days of further negotiations means final terms and their supply chain impact remain uncertain.
Starboard bets AI can give smaller forwarders a fighting chance
Small and mid-sized forwarders face a structural cost disadvantage on the quote desk when competitors automate rate ingestion and carrier portal access. Starboard's premise is worth watching: if AI can handle the messy multi-format data problem at scale, the unit economics for smaller operators improve materially.
CargoWise outage reignites debate over software quality, support and updates
A faulty data update taking down CargoWise for hours exposed how deeply freight operations depend on a single platform. The customer frustration with support response is a prompt for any 3PL or freight manager to verify that manual fallback procedures actually work before the next outage.
FedEx fuel surcharge change raises heat on export shipments
Effective June 22, FedEx export surcharges rise while import rates drop. Shippers with meaningful outbound volume should update landed cost models and check whether routing choices or carrier mix can absorb the differential before the change takes effect.
Nestlé ups automation use with latest distribution center
A 700,000-square-foot DC with the company's largest automated storage and retrieval system, part of a $25B multi-year plan, shows large CPG operators accelerating automation spend even in a soft demand environment. For equipment vendors and integration firms, demand is concentrating in large-program commitments rather than distributed smaller projects.
Deals & Market
5 storiesAI inference startup Baseten reportedly raising $1.5B months after its last mega-round
A second mega-round in months for an inference infrastructure play confirms that compute serving is where venture capital sees durable margin. Enterprises evaluating AI deployment should expect inference pricing to stay competitive as well-funded providers race to scale.
General Intuition in talks to raise $300M at around $2B valuation
Training embodied AI on two billion gaming videos per year is an unconventional data strategy, but if world-model approaches to robotics mature faster than expected, demand for physical automation integrators could shift considerably within the next few years.
BNSF wins local approval for new $4B California rail intermodal project
Local approval for a large intermodal hub near the LA/Long Beach complex is a long-cycle infrastructure bet on sustained import volume. For shippers dependent on West Coast rail drayage, this improves future capacity options, though timelines on projects this size stretch years before they deliver.
Amazon hopes to challenge Nvidia more directly by selling its AI chips
AWS positioning its custom silicon as an external product sold to other data centers signals that Amazon sees chips as a revenue line, not just a cost reduction tool. It adds a credible alternative in the inference hardware market, though Nvidia's software ecosystem advantage remains the harder gap to close.
Cold storage crunch eases as vacancy rates hit 20-year high and supply dries up
Vacancy at historic highs combined with a record-low development pipeline is a classic setup for a rate recovery in 18 to 24 months as current oversupply burns off. Cold-chain tenants with flexibility should look at locking in favorable lease terms before that cycle turns.
Research & Frontier
5 storiesCEO-Bench: Can Agents Play the Long Game?
A benchmark testing whether agents can navigate long horizons under uncertainty, acquire information in noisy environments, and calibrate confidence is more operationally relevant than short-task coding benchmarks. Results will help teams set realistic expectations for autonomous AI agents handling multi-step business processes.
AI systems rival doctors in new Nature studies, but one result suggests the tech won't age well
Two peer-reviewed Nature studies showing AI matches physicians in simulated cases is a credible milestone, but both systems ran on base models that are already outdated at publication. The model churn problem applies directly to any enterprise AI deployment that requires consistent benchmark performance over time.
Google Deepmind treats its own AI agents like rogue employees with office keys
DeepMind tying agent permissions to measurable capability thresholds rather than deployment seniority is a practical governance template for enterprise AI architecture. Scope access proportionally to what an agent can actually do, not to how long it has been running in production.
NAVI-Orbital: First In-Orbit Demonstration of a Zero-Shot Vision-Language Model for Autonomous Earth Observation
A VLM running autonomously on a low-earth-orbit satellite and interpreting imagery without ground-in-the-loop latency is a meaningful step toward continuous monitoring of ports, terminals, and agricultural zones. Near-term supply chain relevance for real-time visibility providers tracking physical asset movements from above.
AI coding agents taught robots how to install GPUs and cut zip ties
Nvidia using AI coding agents to generate robot training tasks rather than hand-authoring them accelerates the pipeline for physical automation skill development. Warehouse robotics integrators should track how fast this approach compresses custom deployment lead times before projecting 2027 and 2028 installation capacity.
Org & AI Architecture
5 storiesGoogle's Gemini co-lead Noam Shazeer joins OpenAI after two-year return stint
The Transformer co-author moving from Google to OpenAI, following Karpathy's move to Anthropic, signals that AI talent gravity has shifted toward labs deploying products at scale rather than pure research organizations. For enterprise buyers, it reinforces OpenAI's technical bench depth heading into its IPO.
Expeditors International to lay off 230 tech workers
A 230-person IT reduction at one of the world's largest freight forwarders is an early data point in tracking how AI adoption actually shows up in logistics headcount. Watch whether Expeditors' operational metrics hold steady as the internal tech workforce shrinks.
Southwest Airlines partners with AWS 'to accelerate AI capabilities and tech modernization'
A full cloud and AI migration targeting 2028 for an airline at Southwest's scale is a multi-year infrastructure commitment, not a pilot program. Competitors watching its on-time and cost-per-seat metrics over the next two years will get the clearest read on whether AI-driven modernization delivers what it promises.
Yann LeCun warns AI labs like OpenAI and Anthropic face a 'big bubble explosion'
LeCun's warning is worth taking seriously even accounting for his stake in an alternative approach. The operational concern for enterprises is concrete: AI vendor funding shocks or consolidation could disrupt tools and APIs that teams have already built workflow dependencies on.
USPS – insourcing is killing the contractors that built its network
3,000 contractor jobs lost in six months as USPS brings delivery operations in-house is a live case study in what concentration risk looks like when a large institutional customer decides to repatriate outsourced work. Any company with high revenue dependence on a single large customer's contracted network should model this scenario.