The Manifest
Toyota is pouring billions into factory robots while OpenAI and Anthropic fight a price war for enterprise customers, all against a fresh warning that most companies outside tech have seen zero financial payoff from their AI spending.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01Toyota is betting $6.4 billion a year on humanoid robots and factory automationThe Decoder
- 02OpenAI offers discounts up to 90 percent to poach Anthropic's biggest customersThe Decoder
- 03Wall Street's fastest traders are building data centers and backing chip startups to stay ahead in AIThe Decoder
- 04AI delivers 'zero financial payoff' to most companies, Apollo economist saysTechCrunch AI
- 05Anthropic says one AI agent beats many subagents now that models can handle complexity aloneThe Decoder
Models & Releases
5 storiesGrok releases the first seven minutes of its AI-generated 'Odyssey'
The footage is serviceable but nowhere near a planning or ops tool, treat it as an entertainment demo rather than a sign of near-term AIGC adoption.
Nanbeige Releases Nanbeige4.3-33B, a 33B Open Agentic Coding Model
Another open-weight coding agent claiming strong terminal benchmarks, worth a look for internal tooling but wait for independent verification outside vendor-reported numbers.
Hugging Face Releases Carbon-X, a 14B Open DNA Language Model
Outside core supply chain relevance directly, but it shows how fast domain-specific LLMs are spreading, pharma and agri procurement teams should keep an eye on this trend.
Nvidia open-sources Flora to make robot video labeling 125x cheaper than GPT-6 Astra
Cuts the cost of labeling robot training data dramatically, which matters directly for anyone piloting warehouse or factory robotics since labeling has been a real bottleneck.
OpenMobile-2 proves open-source smartphone agents can rival proprietary models
Shows open models closing the gap with proprietary agents on real device tasks, relevant if you're weighing vendor lock-in for field or warehouse mobile apps.
Supply Chain & Ops
5 storiesToyota is betting $6.4 billion a year on humanoid robots and factory automation
Toyota is explicitly repositioning itself as a robotics company, not just a carmaker, a signal that factory automation capex is becoming a strategic bet rather than a pilot line item.
Promises vs. reality: Closing the AI adoption gap in transportation
The core point is obvious but true, trust only grows once AI proves reliable in production, and that's the actual bottleneck for most fleet and planning teams right now, not model capability.
How shoppers define delivery success, and what they'll do if you fall short
Roadie's research is a reminder that last-mile misses push customers to competitors fast, useful ammunition for anyone justifying investment in delivery visibility tools.
Navia Deepens Southeast Asia Reach With New Office in Johor Bahru
Freight forwarders keep building out APAC footprint near Singapore's trade lanes, a sign the region's logistics buildout isn't slowing despite broader freight softness.
Borderlands Mexico: Avocado imports generate $641M for US trucking, warehousing
A reminder that produce season still drives real freight and warehouse demand along the border, worth tracking for anyone planning capacity around Mexico cross-border lanes.
Deals & Market
2 storiesOpenAI offers discounts up to 90 percent to poach Anthropic's biggest customers
Buyers should use this price war as leverage in renewal talks, but watch for lock-in costs and support tradeoffs once the introductory discounts end.
Wall Street's fastest traders are building data centers and backing chip startups to stay ahead in AI
When market makers start building their own compute, it tells you GPU access is now a competitive moat even outside tech, expect compute scarcity to persist for everyone else in line.
Research & Frontier
5 storiesAnthropic co-founder Jack Clark moves his prediction for self-improving AI forward to 2027
Worth tracking less for the headline date and more for the underlying benchmark, which measures whether models can design and evaluate their own experiments.
Tsinghua AI Open-Sources AKV, a Cache Engineering Server for LLM Agents on SGLang
A practical fix for agent cost, caching instead of re-summarizing history cuts compute spend by nearly 28 percent, relevant for anyone running agents at scale.
Researchers from Stanford and UC Berkeley Introduce TokenGate: Token-Gated Attention Makes Long-Context Prefill Up to 11.3x Faster
An 11x speedup on long-context prefill without accuracy loss is the kind of efficiency gain that actually lowers the cost of running agents on messy real-world documents.
Permission to Retry: A Study of Authorization and Recovery in AI Agent Workflows
A useful reminder that agent reliability depends on unglamorous plumbing like checkpoints and permissions, not just model quality, exactly the kind of work that decides whether agents are safe in production.
Don't Forget What You Just Learned: Self-Directed Repair Turns Agent Mistakes Into Lasting Lessons
Tackles a real deployment problem, agents repeating the same mistakes because lessons don't persist across sessions, which is exactly the gap between demo and production use.
Org & AI Architecture
4 storiesAI delivers 'zero financial payoff' to most companies, Apollo economist says
A blunt counterpoint to the capex boom, worth sharing with any exec who assumes AI spend alone guarantees ROI outside the tech sector.
Anthropic says one AI agent beats many subagents now that models can handle complexity alone
Anthropic's own guidance to consolidate multi-agent setups into a single agent with on-demand context is a useful signal that orchestration complexity was often overengineering, not a necessary feature.
Ex-Google CEO Eric Schmidt calls OpenAI and Anthropic modern colonial trading companies
The framing is dramatic but the underlying point about capital concentration deciding the AI race matters for any ops leader wondering which vendors will still be standing in five years.
Anthropic Launches OSS Scanner, a Free AI Vulnerability Scanner
Automated scanning without human review before reports reach maintainers speeds up findings but pushes triage work onto already thin open-source teams, worth watching before leaning on it for critical dependencies.