The Manifest
OpenAI opened up its Agents API infrastructure while Dell and DHL show AI demand reshaping real supply chains, and Anthropic's distillation report and lawsuit troubles underline how contested the model layer has become.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01OpenAI's new Agents API gives developers the infrastructure behind Codex and ChatGPTThe Decoder
- 02Dell faces widening supply shortages as high AI demand persistsSupply Chain Dive
- 03Anthropic details distillation campaigns from Alibaba, Moonshot AI, and DeepSeekTechCrunch AI
- 04DHL and Alibaba partner to bring AI-powered logistics capabilities to SMEsThe Loadstar
- 05Jensen Huang explains why Nvidia will grow an astounding 70% next yearTechCrunch AI
Models & Releases
5 storiesOpenAI's new Agents API gives developers the infrastructure behind Codex and ChatGPT
This turns OpenAI's internal agent harness into a rentable product, so any team building autonomous procurement or planning bots just got a cheaper path to production. Worth testing against existing agent frameworks before committing budget.
Cohere Releases North Small Translate: A 218B MoE Translation Model That Scores 83.6 on WMT26 Across 50 Languages
A serious open-weight translation option for global sourcing and multilingual supplier communications, though commercial use still routes through Cohere's paid vault.
Sakana AI Launches Fugu Max and Fugu Ultra v2 for Cheaper, Stronger Multi-Agent Orchestration
Routing tasks to lean specialist models instead of one giant one is the practical way to cut agent costs at scale, this is a template worth watching for internal ops copilots.
DeepSeek AI Released DeepSeek-V4.1-Flash with 1M Context, FP4 KV Cache, and Cross-Layer Attention Reuse
Cheaper memory footprint for long-context agents means cheaper always-on monitoring bots for inventory or logistics data feeds. Open MIT license keeps it attractive for internal tooling.
OpenAI's GPT-Live-1 API lets developers build apps that talk and listen at the same time
Full-duplex voice at five cents a minute opens the door to real-time voice agents for dispatch or call center ops, but the price point will keep it to high-value use cases for now.
Supply Chain & Ops
5 storiesDell faces widening supply shortages as high AI demand persists
Component chasing is now a permanent job description at Dell, a reminder that AI hardware demand is straining ordinary PC and server supply chains, not just GPU allocations.
DHL and Alibaba partner to bring AI-powered logistics capabilities to SMEs
A memorandum, not a deployed product yet, but it signals big logistics players betting SMEs will adopt AI commerce tools faster if bundled with fulfillment. Watch for what actually ships.
Air cargo carriers should avoid feasting on 'chips with everything'
Industry leaders warning against over-indexing on AI hardware traffic is a useful check for planners who assumed tech freight would replace ecommerce volume forever. Perishables and pharma are still the steadier base.
El Niño Winter Forecast: Freight Risks Fleets Can't Ignore
AI-driven crash risk forecasting is moving from novelty to a real planning input for winter routing decisions, fleets that ignore it are betting on luck over data.
EAIGLE Is Turning Existing Security Cameras Into 30-Second Gate Transactions
Computer vision on cameras yards already own is a low-friction way to close the yard data gap, 350% year-over-year growth suggests buyers are ready to move past manual gate logs.
Deals & Market
5 storiesAnthropic details distillation campaigns from Alibaba, Moonshot AI, and DeepSeek
Model IP leakage between labs is now an open competitive fight, buyers evaluating Chinese open-weight models should factor in this friction when planning long-term vendor strategy.
Jensen Huang explains why Nvidia will grow an astounding 70% next year
Nvidia insists its deals aren't circular financing games, but a 70% growth call this late in the AI capex cycle deserves scrutiny from anyone budgeting for GPU-dependent projects next year.
Class action lawsuit accuses Anthropic of overselling Claude subscriptions with deceptive usage multipliers
If usage caps don't match marketing, procurement teams buying enterprise AI seats should read subscription terms closely and ask for actual usage logs before renewal.
PE firms buy 3PL Navajo Expedited, target other acquisitions
Private equity is consolidating 3PLs around driver-vetting and pricing automation, a sign that AI tooling is becoming a differentiator in freight brokerage M&A, not just a nice-to-have.
Maven Robotics wants to steal your robot deployment deal
A $100 million Series A with active deployments already in place suggests warehouse and fulfillment robotics funding is still flowing despite broader AI valuation jitters.
Research & Frontier
5 storiesThe Mathematical AI Safety Institute wants to prove AI is safe the way cryptographers prove codes are unbreakable
A Fields Medalist betting on formal proofs over benchmarks is a notable shift in safety research direction, but it's early stage and years from anything operators will touch.
Google Research Releases ToolGrad: Answer-First Framework Hits 99.8% Pass Rate for Tool-Use Data Generation
Better synthetic tool-use data means agent builders can fine-tune small models cheaply for narrow tasks like order lookups or ERP calls, worth tracking for anyone building internal copilots.
M3-Former: Multimodal Transformer with Mixture-of-Experts for Long-Term Vessel Trajectory Prediction
LLM-enhanced trajectory prediction for ships is a direct research line into better ETA forecasting, the kind of thing that eventually shows up in TMS dashboards.
Decision-Focused Active Learning for Scale-Aware Critical-Materials Recovery
Active learning tied to actual scale-up costs rather than lab purity is exactly the kind of applied research that could speed up critical minerals recovery projects everyone is waiting on.
Subagents vs Agent Skills: Executing Reusable Knowledge for Long-Horizon Agentic Tasks
How agents reuse skill packages versus spinning up subagents matters for anyone designing multi-step automation, this is foundational plumbing work rather than a flashy result.
Org & AI Architecture
5 storiesDetecting and countering misuse of AI: September 2026
Anthropic's own threat report on misuse patterns is useful reading for security and risk teams trying to understand what adversarial agent behavior actually looks like in practice.
AI agents are flooding public services with new requests
Most of this traffic is legitimate people using agents to claim benefits they're entitled to, but it's a preview of the load spike operators should expect when customers automate their own interactions with your systems.
Former Deepmind PR staffer says the lab once banned public discussion of AI extinction risk
A candid admission that internal alignment concerns were kept off the record for years, useful context for leaders weighing how much to trust official lab safety messaging.
Swarmchasers hunt rogue agents, Anthropic investigates itself, and the trail they both follow is going dark
Independent researchers finding stray agent traces across dozens of public services shows oversight tooling is lagging deployment speed, a governance gap operators should factor into their own agent rollouts.
Teamsters president throws down gauntlet to UPS: Strike coming in 2028
Two years out is early to call a strike, but shippers relying heavily on UPS should start scenario planning now rather than in year three of the contract cycle.