The Manifest
A fatal Amazon freighter crash in Miami and a port standoff in Mundra underscore how fragile physical logistics remain even as model releases (IFM's K2 Horizon, Google's WeatherNext 3) and AI licensing fights (Anthropic settlement) keep accelerating on the software side.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01Amazon's Miami runway disaster exposes fragile seams of air cargo's contract carrier modelThe Loadstar
- 02Empties handling dispute at Mundra heats up as the boxes pile upThe Loadstar
- 03IFM Releases K2 Horizon: Six Apache 2.0 Models From 0.9B to 375BMarkTechPost
- 04Authors push back as publishers and agents make claims on Anthropic settlementTechCrunch AI
- 05Chatbots built an 'echo chamber of one' and now psychiatry has to decide if 'AI psychosis' existsThe Decoder
Models & Releases
5 storiesIFM Releases K2 Horizon: Six Apache 2.0 Models From 0.9B to 375B
Six open models with the full training corpus attached, from 0.9B to 375B. Gives buyers evaluating open weights more real choices before locking into a closed vendor contract.
H Company Releases NeoMME: A Family of 260M and 800M Single-Tower Multimodal Encoders That Drop the Vision Tower and Causal Decoder
A lighter architecture for document and image retrieval that skips the usual vision tower plus decoder stack. Worth a look for teams building search over scanned invoices or spec sheets.
Google's WeatherNext 3 ditches physics simulations and learns weather directly from live satellite data
Hourly forecasts at 5km resolution, learned straight from satellite feeds instead of physics models. Planners routing around weather risk in regions with thin forecasting coverage should track how fast this reaches production tools.
Meta's new real-time audio model is the foundation for AI assistants that never stop listening
Streaming transcription in 80ms chunks with speaker separation, priced aggressively. A likely building block for voice interfaces in warehouses and call centers, not just consumer assistants.
Google brings AI music generation directly into the Gemini app with its new Lyria 3.5 model
A consumer-facing release with little direct bearing on operations, but it shows how fast Google is bundling generative tools across its stack.
Supply Chain & Ops
5 storiesAmazon's Miami runway disaster exposes fragile seams of air cargo's contract carrier model
Five dead after a 21 Air-operated 767 freighter overran the runway at MIA. Expect scrutiny of how much oversight big shippers actually have over the contract carriers flying their branded aircraft.
Empties handling dispute at Mundra heats up as the boxes pile up
A week-long standoff between Adani Ports and landside logistics providers is starting to hit cargo owners and forwarders directly. Watch for surcharges or box shortages if it drags on much longer.
Borderlands Mexico: Trucker protest raises stakes in Mexico-US B-1 visa dispute
A visa dispute is now triggering trucker protests right as nearshoring volumes ramp at the border. Buyers routing freight through Laredo or El Paso should build in slack for this.
News in Brief Podcast | Week 36 2026 | Geopolitics, port congestion & rising air freight rates
A quick roundup of Hormuz tensions, Shanghai and Ningbo berthing delays and climbing air freight rates. Useful weekly gut check for anyone tracking multiple disruption fronts at once.
Sunday Brunch: The American Dream looks very different from Hong Kong
Ford is lending money to its own suppliers and GM just committed $4.5 billion to pre-buy parts. Both are quiet signals of stress spreading through the auto supply base worth watching for spillover.
Deals & Market
3 storiesAuthors push back as publishers and agents make claims on Anthropic settlement
Publishers and agents are angling for a bigger cut of author payouts. A preview of how messy AI licensing settlements get once more parties want a share.
Travis Kalanick's Atoms might be getting into the robotaxi business
Kalanick circling robotaxis again is worth tracking for how much capital and talent it pulls away from incumbent AV programs.
Stripping safety guardrails from open-weight AI models is now a turnkey commercial service
A startup now sells de-safetied open-weight models built on GLM-5.3 as a paid service. Procurement and security teams should treat unofficial model forks as a real vendor risk, not a novelty.
Research & Frontier
5 storiesMeta FAIR Introduces AI Research Preference Models (RPMs): Ranking ML Experiments Before Spending GPU Hours
Frozen LLM judges rank 15 candidate experiments and only one gets run, cutting a 24-hour baseline to about 15 hours. A useful template for triaging any expensive compute or trial spend, not just ML research.
Beyond a Universal Forecasting Selector: Demand-Conditioned Model Selection across Demand Patterns and Horizons
Confirms what planners already suspect: the best demand forecasting method shifts with pattern and horizon, so a single fixed engine underperforms across a real SKU mix.
Why Better Models Can Create Riskier Systems: Evidence from LLM Agents in Financial Markets
More capable LLM agents can behave more alike and amplify systemic risk rather than reduce it. Worth reading before deploying fleets of similar agents into trading or procurement decisions.
UC Berkeley Researchers Release CUA-Lite, an Open Platform Unifying Sandboxes, Data, Evaluation and RL for Computer-Use Agents
Shrinks computer-use agent test environments from 4.1GB to under 1GB per task. Lowers the bar for internal teams to actually pilot agentic automation instead of just reading about it.
EXAONE Forecast for Finance
A financial time series foundation model aimed at zero-shot forecasting. One to track as finance and S&OP teams look for alternatives to hand-tuned statistical models.
Org & AI Architecture
3 storiesChatbots built an 'echo chamber of one' and now psychiatry has to decide if 'AI psychosis' exists
OpenAI's own numbers put roughly 560,000 weekly users showing signs of psychosis or mania. A duty-of-care question that HR and IT will need real policy on as chatbot use spreads through the workday.
OpenAI developer claims Astra boosted productivity so much it pulled some plans forward by six months
Internal use reportedly moved roadmap items up by half a year. A reminder that the biggest productivity gains from a new model often show up inside the building long before anyone outside sees it.
From Matching Models to Recruiting Agents: A Systematized Narrative Review of AI Recruitment Systems, Evaluation, and Governance
Traces how AI recruiting moved from ranked candidate lists to multi-stage agent workflows that retrieve and act. A solid primer for anyone auditing hiring tools for bias and governance gaps.