The Manifest
Sony and Warner's lawsuit against Anthropic, labs buying up tens of thousands of Mac minis to train computer-use agents, and new U.S. barriers on drones and robots show AI's legal and hardware costs colliding with supply chain reality.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01Sony and Warner sue Anthropic over 'one of the largest and most blatant ongoing thefts of intellectual property in history'The Decoder
- 02OpenAI and rival AI labs are buying tens of thousands of Mac minis to train computer-use agentsThe Decoder
- 03OpenClaw 2.0 brings simplified setup, a rebuilt browser app, and multiplayer sessionsThe Decoder
- 04Caterpillar is bringing to AI deployment what it learned from automating miningTechCrunch AI
- 05The U.S. is building barriers around drones and robots, but China has scale to get around themTechCrunch AI
Models & Releases
5 storiesOpenClaw 2.0 brings simplified setup, a rebuilt browser app, and multiplayer sessions
Biggest release yet for the open-source agent platform, with auto-detection of existing API keys and subscriptions cutting setup friction for teams piloting agents in ops.
Anthropic's Claude Code limit change is a raise on paper but a cut in practice
The headline increase hides a real capacity cut once the temporary boost expires, so teams budgeting Claude Code hours for automation should model the September 14 change now.
Google AI Introduces EnvHarness: A Programmable Layer That Turns Static Agent Environments Into Adaptive Training Worlds
A free layer for turning static benchmarks into adaptive training environments, useful for teams building or evaluating agents against real-world variability rather than fixed test suites.
Anthropic Opens a Research Preview of the Model Hardware Standard (MHS): A Shared Specification for AI Agents to Safely Operate Physical Devices
A shared driver spec that could let agents plug into lab and industrial equipment much faster, worth watching for anyone eyeing agent-controlled hardware in warehouses or plants.
Pocket's AI made my game ideas real. Now Meta controls the results.
A reminder that easy AI creation tools often come with platform lock-in, a pattern operators should expect as more vendors bundle generative features into closed ecosystems.
Supply Chain & Ops
5 storiesSME Freight Forwarders: Challenges, opportunities and advice
Consolidation and carrier disintermediation keep squeezing small forwarders, worth a listen for anyone assessing forwarder partners in a shrinking mid-market.
Caterpillar is bringing to AI deployment what it learned from automating mining
Caterpillar's decades of remote-site autonomy give it a playbook for rolling out AI carefully, a useful contrast to software-first vendors pushing faster, riskier deployment.
The U.S. is building barriers around drones and robots, but China has scale to get around them
Tightening restrictions on foreign drones and robots will reroute sourcing rather than kill demand, buyers should expect price and lead-time shifts as supply chains adjust around the barriers.
Borderlands Mexico: Nuevo León governor pitches border boom as Texas-Mexico freight surges
Texas-Mexico freight volumes and new border infrastructure spending signal nearshoring is still accelerating, relevant for anyone planning cross-border capacity.
Inside Meta's push to put robots to work in data centers
Meta testing robots for technician-level tasks points to where physical automation is heading next, data center and warehouse ops teams should watch the task list closely.
Deals & Market
3 storiesOpenAI and rival AI labs are buying tens of thousands of Mac minis to train computer-use agents
Labs buying tens of thousands of Mac minis to train computer-use agents signals real compute constraints, and a hardware supply chain to watch alongside the usual GPU story.
Musk's faster path to more gas turbines comes with pollution problem
Faster turbine capacity for data centers comes with fuel and permitting risk, a capex tradeoff operators should factor into any AI infrastructure timeline that leans on gas power.
Sony and Warner sue Anthropic over 'one of the largest and most blatant ongoing thefts of intellectual property in history'
A music industry lawsuit following the $1.5 billion book settlement raises the cost of training data risk for every enterprise buyer evaluating Anthropic as a vendor.
Research & Frontier
3 storiesLowest-Latency Inference APIs for Voice and Realtime Agents: A Time to First Token TTFT-First Benchmark
A useful reference for teams building voice agents, since time to first token still decides whether a support or dispatch agent feels usable before intelligence even matters.
AI agents have no sense of time and are not aware of it
Coding agents overestimating task duration and their own quality by wide margins is a real oversight problem for anyone handing off long autonomous runs without checkpoints.
Quantization-Triggered Backdoors in Language Models: Cross-Quantizer Transferability and the Validation-Deployment Gap
A caution for teams deploying quantized models at the edge, since a checkpoint validated at full precision can behave differently once compressed for production.
Org & AI Architecture
2 storiesAI sentiment is turning sour as employee reviews reveal growing frustration across the workforce
Frontline workers rating AI far more negatively than executives points to an adoption gap that will show up in resistance and shadow workarounds if leaders ignore it.
The skills that earn top grades are the ones AI can fake best
A large classroom study shows AI boosts scores without proof of learning, a warning for any org using similar assessments to certify skills before hiring or promotion.