The Manifest
OpenAI shipped its three-tier GPT-5.6 family and a new workflow agent the same day Fidji Simo stepped down from its No. 2 role, while container carriers keep bracing for Red Sea-driven capacity shocks and freight tech consolidation.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01OpenAI launches its new family of models with GPT-5.6TechCrunch AI
- 02Fidji Simo steps down from OpenAI's no. 2 roleTechCrunch AI
- 03Meta enters the crowded AI coding battle with Muse Spark 1.1TechCrunch AI
- 04EXCLUSIVE: MSC vs CMA CGM - twists and turns at Clasquin & CevaThe Loadstar
- 05Ford, GM sign memory supply agreements with MicronSupply Chain Dive
Models & Releases
5 storiesOpenAI pairs its GPT-5.6 public rollout with ChatGPT Work, a new agent that handles entire workflows
The agent runs projects across Drive, Slack and Salesforce on its own. Worth a pilot for teams drowning in cross-app busywork, but check what happens when it hits an exception.
Grok 4.5 is so cheap compared to Fable 5 and GPT 5.5 that benchmark gaps may not matter much
It trails the top models on coding benchmarks but uses far fewer tokens per task. For teams running high-volume agentic workloads, the price per task matters more than the leaderboard spot.
Google Research Introduces SensorFM: A Wearable Health Foundation Model Pretrained on One Trillion Minutes of Sensor Data
Massive scale pretraining on wearable sensor data. Not a supply chain tool today, but the same approach could eventually feed fatigue and safety monitoring for warehouse and driver workforces.
Meet Nemotron Labs 3 Puzzle 75B A9B: A Compressed Hybrid MoE LLM Delivering 2.03x Server Throughput
Compression that roughly doubles throughput without a full retrain. Relevant for anyone running inference at scale who wants to cut GPU spend without swapping models.
Meet LingBot-World-Infinity: An Open Causal World Model With An Agentic Harness
An open-weight interactive world simulator out of Ant Group's robotics unit. Worth watching for teams testing warehouse or fulfillment automation in simulation before touching real hardware.
Supply Chain & Ops
5 storiesForwarders fear for time-critical cargo as Kenya plans new security measures
A proposed export-control bill could add licensing steps that slow aircraft spare parts clearance. Forwarders moving AOG shipments through Kenya should track the bill's timeline now, not after it passes.
Scotts Miracle-Gro widens tech partnership for supply chain AI
Another brand consolidating planning, decisioning and execution onto one Kinaxis platform instead of stitching together point solutions. The trend toward single-platform planning keeps building.
Amazon Shipping looks to snag FedEx, UPS customers with low pricing
Fewer surcharges and aggressive rates now that the service is open to all businesses. Shippers with volume should be quoting Amazon against incumbent carriers this quarter.
FedEx launches 'dedicated life sciences organization'
A named healthcare and life sciences unit signals carriers still see cold-chain and pharma logistics as the growth lane worth building specialized teams around.
What supply chain visibility gets wrong about ETA numbers and why it matters
The industry's go-to '90% accuracy' figure doesn't hold up under scrutiny across 25 interviews with shippers, carriers and vendors. Anyone buying a visibility platform on that stat should ask what exactly it's measuring.
Deals & Market
5 storiesAn AI agent startup just let its agent run its $100M fundraise
Lyzr's agent reportedly managed parts of its own $100M round. Cute proof of concept, but investors and buyers should still ask what a human verified before wiring money.
Paris-based AI voice startup Gradium raises $100M seed, backed by Nvidia
A $100M seed round for a voice AI startup shows mega-round sizing hasn't cooled even in narrower product categories.
Elon Musk praises Mythos/Fable, promises not to 'cut off' Anthropic
With roughly $40 billion in Anthropic revenue riding on compute access, the promise is only as good as the relationship. Anyone evaluating vendor risk should note the concentration here.
Can AI answer the $3 trillion question?
The ROI debate is back with bigger dollar figures attached. Buyers renewing AI contracts this year should push vendors past the topline productivity claims for actual unit economics.
Databricks makes Chinese open-source model GLM 5.2 its default coding engine after it matched Opus at lower cost
Databricks benchmarked on its own multi-million-line codebase and picked the cheaper open model over Opus. The lesson for buyers is to run your own tests instead of trusting public leaderboards.
Research & Frontier
5 storiesContext Graphs for Proactive Enterprise Agents
Argues enterprise agents need to surface information before someone asks for it, not just answer queries. Relevant framing for anyone building exception-management tools on top of ERP data.
AI-integrated models for assessing agricultural resilience
Couples economic (GTAP) and crop (APSIM) models so policymakers can query supply chain shock impacts directly. A useful template for scenario planning beyond just agriculture.
Humanoid robots controlled by surgeons did world-first operation on live pigs
A preclinical trial of remote humanoid-operated surgery. Early, but it's a real data point on remote robotic control moving beyond fixed surgical arms.
OpenAI finds roughly 30 percent of popular AI coding test is broken
OpenAI pulled its endorsement of SWE-Bench Pro after finding nearly a third of tasks flawed. A good reminder to treat vendor benchmark claims with a healthy dose of skepticism.
OpenAI's AI beats every human at AtCoder, a top competitive programming contest
A system solved all five Algorithm Division problems, including two rated exceptionally hard. Impressive on paper, though the gap between contest performance and messy production code remains wide.
Org & AI Architecture
4 storiesNew York Times says OpenAI hid evidence in ChatGPT copyright trial
Publishers claim OpenAI hid tools and datasets and are pushing for sanctions. Any enterprise leaning on ChatGPT outputs for content work should watch how this discovery fight lands.
Google will now disclose which ads are made with AI
Disclosure requirements expand beyond election ads to general AI-generated or altered content. Marketing and comms teams should get labeling practices in order before this becomes enforced more broadly.
Instagram users: Here's how to stop Meta's AI from using your photos
Public profile photos get swept into Meta's image generator by default unless a user opts out. A small story with a big message: assume any public content is training or generation fodder now.
How did the government decide OpenAI's frontier model was safe to release?
The exact dialogue between regulators and frontier labs before a release remains opaque. Worth watching for anyone in a regulated industry that has to justify its own AI vendor choices to auditors.