The Manifest
Google and Anthropic push deeper into enterprise agents while OpenAI faces scrutiny over its revenue numbers and a mathematician backlash, and AI-driven data center demand keeps reshaping trans-Pacific freight just as Q4 customs and capacity pressure builds.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01Google brings agentic AI to Gemini, starting with businessesTechCrunch AI
- 02OpenAI's revenue is reportedly $20 billion less than previously projectedTechCrunch AI
- 03AI gold rush transforms trans-Pacific trade and logisticsFreightWaves
- 04One public-facing AI agent on AWS could read, rewrite, and delete every other agent in the regionThe Decoder
- 05Some mathematicians call for OpenAI boycott after AI-generated proofs flood their fieldThe Decoder
Models & Releases
5 storiesGoogle brings agentic AI to Gemini, starting with businesses
Google is pushing a single agent that can delegate to subagents and touch enterprise systems directly, worth watching for anyone evaluating procurement or planning copilots since it shows where big platform vendors expect agents to live.
Claude Haiku 5.5 arrives with massive price cuts proving the AI pricing arms race is far from over
Up to 90 percent lower token prices and a big jump on computer-use benchmarks make Haiku a credible option for high-volume, low-stakes automation like document parsing or data entry in ops workflows.
JetBrains Releases Mellum2.1: A 12B MoE Open Model for Coding Agents
An open, Apache-licensed coding model with a big SWE-bench jump is good news for teams building internal tools rather than paying for proprietary coding agents.
Claude can now generate animated explainer videos and live data dashboards from text prompts
Turning BigQuery or Snowflake data into live dashboards from a text prompt could cut the busywork of building reporting views, but check the numbers before trusting them in a planning meeting.
Nvidia's big bet on physical AI aims for safer robotaxis, humanoid robots
A full-stack safety layer for robotaxis and humanoid robots matters for anyone tracking when warehouse and yard robotics get cleared for wider deployment.
Supply Chain & Ops
5 storiesAI gold rush transforms trans-Pacific trade and logistics
Data center buildouts are now a real freight demand driver in their own right, reshaping capacity and pricing on trans-Pacific lanes independent of the usual consumer goods cycle.
WiseTech AI charge plan raises questions over CargoWise Value Packs
CargoWise customers should check their contracts now, since AI agent features look set to be billed outside existing Value Packs and that changes the ROI math on automation.
Customs updates add a 'compliance burden' to potentially volatile Q4 for shippers
New US, Mexico, and EU customs rules land right before peak season, adding paperwork risk on top of the usual Q4 volume crunch, worth a dry run with brokers before November.
Global container volumes hit record 17.46 million TEUs
A record month with pricing at a two-year high confirms demand is outrunning easy capacity, so expect spot rate volatility to persist into Q4.
Factory backlogs build as freight and steel costs climb
Unfilled orders hit $1.61 trillion while shipments stalled, a sign that working through backlogs is getting more expensive even as volume growth slows.
Deals & Market
5 storiesOpenAI's revenue is reportedly $20 billion less than previously projected
If the gap between reported and actual revenue is this wide, enterprise buyers should press harder on usage data before signing multi-year AI commitments tied to any single vendor's growth story.
Popular AI leaderboard Arena nearly doubles valuation to $3.1B in 10 months
A near-doubling in ten months shows how much money is chasing model evaluation infrastructure, and the pivot toward alignment scoring suggests benchmarks are becoming a trust layer, not just a leaderboard.
Trump unveils plans to build $15B steel mill in Iowa
A $15 billion mill plus a $3 billion iron ore mine adds real domestic steel capacity, relevant for anyone pricing manufacturing inputs or betting on reshoring timelines.
CBP starts paying out IEEPA tariff refunds for finally liquidated entries
Refunds are only flowing to importers who filed a valid importer of record number by July 30, so check eligibility now rather than assuming a blanket payout.
Molson Coors taps chief supply chain officer for Americas
Another large CPG brings in a dedicated Americas supply chain chief, a reminder that modernization mandates are becoming a standard executive job description, not a side project.
Research & Frontier
5 storiesOne public-facing AI agent on AWS could read, rewrite, and delete every other agent in the region
A single public-facing Bedrock agent could hijack every other agent in the account, a sobering reminder to audit default permissions before rolling out fleets of internal agents.
Synthesis Through Simulation: Generating Coherent Enterprise Data via Scalable Agent-System Interaction
A useful approach for teams trying to train or evaluate tool-calling agents on realistic business data without exposing actual enterprise systems or schemas.
The Harness as the Only Mutable Surface: Compliance-Bounded Self-Evolution of LLM Agents in Credit Pipelines
A practical governance pattern for anyone letting agents touch regulated workflows, keep self-improvement confined to a reviewable harness rather than the model itself.
Agent-Controlled Forgetting for Tool-Using Agents: Reversible Context Curation in Practice
A method for agents to compress old tool outputs while keeping a recoverable archive, relevant for anyone hitting context limits on long-running planning or sourcing agents.
Goodfire says its new 'inside-out' monitors catch rogue AI agents at a fraction of the cost
Cheaper internal monitoring could make it more affordable to run safety checks on agent fleets at scale instead of paying a second model to audit every action.
Org & AI Architecture
5 storiesSome mathematicians call for OpenAI boycott after AI-generated proofs flood their field
Terence Tao's warning that AI is harvesting open problems faster than the field can verify them is a preview of the same quality control fight that will hit any domain flooded with AI generated output, including forecasting and planning.
Fired OpenAI safety researchers dispute misconduct claims, warn of chilling effect
Whatever the merits of this specific case, a public dispute like this makes it harder for safety teams anywhere to raise concerns internally without fear of retaliation.
Anthropic changes usage policy to ban model abuse and election interference
Banning sustained abuse of a model and tightening rules on election interference and surveillance signals vendors are getting more explicit about acceptable use, worth checking against your own internal AI policy.
Customer demands for new tech a dilemma for SME forwarders
Smaller forwarders are being pushed to adopt AI and visibility tools their customers may never see, squeezing the personal service edge that used to set them apart from the giants.
AI math breakthroughs have Ethereum researchers debating how fast wallet security could collapse
No one has broken ECDSA yet, but the debate over migration timing is a useful case study in how to plan for a low probability, high impact technology risk without overreacting.