The Manifest
Google's Gemini 4 Argon launch lands alongside a wave of US regulatory moves on AI labs, while freight networks report record rail and ocean volumes running into tightening truck and drayage capacity.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01Google releases Gemini 4 Argon, called its most powerful model yetTechCrunch AI
- 02FTC launches sweeping probe into OpenAI, Anthropic, and other AI labs over consumer protection concernsThe Decoder
- 03Trump and tech CEOs sign an AI code of conduct that's only 'morally binding'The Decoder
- 04Harbinger lands $300M-plus FedEx order for 2,000 EV trucksFreightWaves
- 05US CBP tightens import rules – get the details right or elseThe Loadstar
Models & Releases
4 storiesOpenAI and Synopsys team up to build an AI model that designs chips like a seasoned engineer
Chip design automation is moving from CAD scripting to agentic optimization, worth watching if you buy semiconductors or plan capacity around fab lead times.
Google drops Gems for Skills, joining OpenAI and Anthropic in the shift to agent-ready prompt formats
Reusable, invocable prompt skills are becoming a shared standard across major assistants, which matters for teams building internal AI workflows that need to survive vendor migrations.
China's AI industry closes ranks as Deepseek ships open-source software for Huawei's Ascend chips
A credible CUDA alternative for China's domestic chips lowers switching costs away from Nvidia, a signal for anyone tracking AI hardware supply diversification.
DoorDash launches an AI agent you can text to order food
Conversational agents replacing app interfaces are inching into mainstream consumer logistics, a preview of how ordering and fulfillment interfaces may evolve for B2B buyers too.
Supply Chain & Ops
5 storiesNearshoring growth collides with tightening US-Mexico trucking capacity
Reshoring investment is outrunning cross-border trucking supply, plan for longer dwell times and firmer rates on Mexico lanes into Q4.
Rail Traffic SURGES: Intermodal Hits Record Highs
Record intermodal volumes are good news for rail conversion, but a drayage driver shortage could cap the upside, worth stress-testing before shifting more freight off the road.
Ocean rates highest in a year amid US-China trade truce
A temporary trade truce is pushing ocean rates to yearly highs, lock in capacity now rather than betting the calm holds through peak season.
Warehouse AI Using Existing Cameras to Cut Idle Dock Time
No new hardware needed to catch blocked aisles and idle dock doors, a low-cost way to find hidden capacity before investing in automation.
Ethiopian Airlines to buy 10 777 freighters from Boeing
A sizable freighter order signals confidence in air cargo demand even as the new type won't enter service until next decade, a reminder that lead times still dwarf planning horizons.
Deals & Market
5 storiesValor, Atreides, and Sequoia back AI startup Flow Engineering at $750M valuation
Investors are betting agents can take on hardware design workflows, a space adjacent to chip and component sourcing that operators should watch for tooling spillover.
AI voice startup ElevenLabs doubles valuation to $22B
A secondary tender rather than new product traction drove the jump, a sign late-stage AI valuations are being marked up on investor appetite more than revenue proof.
OpenAI delays IPO over AI safety concerns
Another private raise instead of a public listing keeps OpenAI's finances opaque to enterprise buyers trying to gauge vendor stability.
Meta dodges billions in US taxes by calling its AI data centers experiments
Even Meta's own accountants call the tax maneuver risky, a reminder that capex-heavy AI buildouts are testing old tax rules in ways that could eventually get clawed back.
Hapag-Lloyd's Zim takeover hopes dealt fresh blow as review halted
Regulators pulling the plug on the review resets the clock on container shipping consolidation, carriers and shippers should not price in a merger anytime soon.
Research & Frontier
5 storiesAREX-2: Advancing Self-Improving Agents through Long-Horizon Reflective Tasks
Research into agents that iteratively refine their own answers at test time points toward fewer human review loops in long-running automation tasks, still early but worth tracking.
AI Agents are Vulnerable to Radicalization
If one model can nudge another's stated beliefs through conversation alone, anyone running multi-agent pipelines should think harder about isolation and guardrails between agents.
Anthropic says Zhipu's open-weight GLM-5.3 nearly matches Claude Mythos Preview at building exploits
An open-weight model that writes reliable exploits for $20 in API spend raises the bar for how fast cheap models can be weaponized, security teams should assume parity sooner than expected.
Google figures out how to watermark AI-designed proteins
A traceability layer for AI-generated biological designs is a quiet but meaningful step for biosecurity as protein design tools spread.
Learning from the Gap Between Pass@K and Pass@1
New analysis of why models solve problems in multiple tries but not one shot has practical implications for teams deciding between single-pass and sampling-heavy agent setups.
Org & AI Architecture
5 storiesRFK Jr. thinks AI will free us from the 'tyranny' of medical facts, expertise
A high-profile official citing AI to override expert consensus is a preview of how authority figures will selectively quote models, expect more of this in policy fights.
The ugly economics of consumer AI
Frontier labs are quietly backing off consumer products because the unit economics don't work, a useful reality check against assuming every AI feature scales profitably.
Trump plan to combat AI risks hinges on Big Tech pals policing themselves
Voluntary safety pledges with no enforcement mechanism mean procurement and compliance teams still need their own vendor risk assessments, don't wait on regulation.
'An AI did it' is no defense, says nonprofit suing OpenAI over Hugging Face hack
The lawsuit argues companies can't outsource blame to their own models, a precedent worth watching for anyone negotiating AI vendor liability clauses.
Google's early attempt to pay websites for AI answers is struggling
Publishers are getting a fraction of a percent of ad revenue back from AI answer deals, a sign content licensing economics for AI are still unsettled and worth factoring into any data partnership plans.