Daily AI News | Anthropic opens a research preview of personal agent Claude Duet
# Daily AI News | Anthropic opens a research preview of personal agent Claude Duet
## Lead Story
### Anthropic opens a research preview of personal agent Claude Duet
Anthropic has opened a research preview of its personal agent Claude Duet. Reports describe proactive reminders and task execution; eligibility, regions and connector counts should be stated only as specified in Anthropic’s announcement.
**AI take:** Persistent task execution moves assistants beyond question-and-answer use, making permission boundaries and user control central.
## More News
### DeepSeek open-sources DeepInfer, its production inference framework
SiliconANGLE reports that DeepSeek open-sourced DeepInfer, identifying it as the inference framework behind the company’s production services.
**AI take:** Opening production infrastructure gives developers a way to inspect and adapt the implementation.
### NVIDIA releases the OpenShell reference architecture for agents
NVIDIA released the OpenShell reference architecture for building, containing, observing and enforcing policies on agents across enterprise infrastructure and edge systems.
**AI take:** Isolation and behavior monitoring are becoming part of the infrastructure choices organizations make when deploying agents.
### European partners adopt NVIDIA DSX for more water-efficient AI factories
NVIDIA says European partners are adopting its DSX AI Factory Platform to build facilities designed to use less cooling water and power for the same AI output.
**AI take:** As AI compute expands, energy and water efficiency are becoming practical factors in infrastructure planning.
### NVIDIA releases RFD4-Proteina and expands healthcare research collaborations
NVIDIA announced collaborations with healthcare and research organizations on protein design, drug discovery and cancer care, alongside the release of its open RFD4-Proteina model.
**AI take:** Pairing an open model with research partnerships points to deeper use of generative AI in life-science workflows.
### NVIDIA expands Isaac robotics with a humanoid reference and new data
NVIDIA expanded its open Isaac robotics platform with a humanoid reference design, datasets and runtime, along with updates to its robot-learning framework.
**AI take:** Combining simulation, data and runtime tools can help teams bridge the gap between robotics experiments and deployment.
### NVIDIA expands open models across language, robotics, biology and quantum
NVIDIA says its open-model family is expanding across language, robotics, biology and quantum computing, including new Nemotron models.
**AI take:** Open models may help researchers adapt systems to local needs.
### NVIDIA introduces modular DSX PODs for faster AI-factory deployment
NVIDIA introduced DSX POD, a modular reference design for Vera Rubin AI factories. The company says deployment at a prepared site could fall from about six months to as little as 14 days.
**AI take:** Pre-integrating power, cooling and networking could change project timelines if the stated deployment targets hold.
### NVIDIA adds Confidential Bare Metal for sensitive model workloads
NVIDIA introduced Confidential Bare Metal, a deployment option for frontier-model builders to run sensitive workloads on dedicated third-party infrastructure while retaining control of models and data.
**AI take:** The dedicated deployment model gives teams with isolation requirements another infrastructure option.
### NVIDIA cuDSS library reaches general availability for simulations
NVIDIA made its CUDA-X cuDSS library generally available for moving critical parts of engineering and scientific simulations onto GPUs.
**AI take:** General availability lets engineering teams evaluate GPU acceleration in production simulation workflows.
### NVIDIA details European Hyperion programs with automakers and mobility firms
At GTC Berlin, NVIDIA announced European autonomous-mobility programs involving partners including Mercedes-Benz, Volkswagen Commercial Vehicles and Einride; Volvo Cars is adopting Hyperion and Alpamayo software.
**AI take:** These are announced partnerships and adoption plans; broad road deployment remains a separate step.
### Telenor AI Factory selects F5 for its sovereign-AI service plane
F5 says Telenor AI Factory selected BIG-IP Next for Kubernetes as the service plane for its NVIDIA-powered sovereign-AI infrastructure.
**AI take:** Sovereign AI also depends on its networking layer.
### Cadence makes a sovereign AI design platform available in Munich
Cadence announced availability of a sovereign-AI data-center platform in Munich combining its design software with NVIDIA systems and digital-twin tools, with European data residency.
**AI take:** Combining design software and compute under regional data controls may serve chip teams with residency requirements.
### NVIDIA and partners plan to expand AI-infrastructure manufacturing
NVIDIA says it and supply-chain partners plan to expand global manufacturing capacity to meet demand for accelerated computing and its AI-factory platform.
**AI take:** The pace of compute buildout also depends on manufacturing, supply and delivery capacity.
### Glean brings headless agent building to Claude Code, Cursor and Codex
Glean made headless agent building generally available, letting developers create, test and publish agents as code from environments including Claude Code, Cursor and Codex, with agents running on Glean’s managed runtime.
**AI take:** Putting agent creation into coding workflows may make automation easier for teams to test and reuse.
### Walmart launches an Agent Gateway for AI-assisted shopping
TechCrunch reports Walmart launched an Agent Gateway for AI agents shopping on customers’ behalf, with spending and transaction-approval controls.
**AI take:** Agent shopping will test payment controls and consumer safeguards.
### Channel Talk introduces its Voice ALF AI phone service
Channel Corporation announced Voice ALF for Channel Talk, a phone service that supports voice-based AI customer consultations.
**AI take:** Escalation and accuracy will shape its customer-service value.
### AiMOGA opens global pre-orders for its first home companion robot
AiMOGA Robotics says global pre-orders opened for Mobo, its first home companion robot, marking a move from commercial services into consumer home robotics.
**AI take:** Consumer robotics is moving from demonstrations toward tests of price, reliability and usefulness in homes.
### Plaud says its devices have surpassed three million users worldwide
CoRRiENTE reports Plaud’s product line exceeded three million cumulative users as of September 30, with more than 1,000 retail locations in Japan and over 16,000 organizational deployments.
**AI take:** The figures indicate adoption of AI voice devices, but do not establish active use or retention for any single product.
### Naver D2SF invests in on-device AI startup TraceWeave
Yonhap reports Naver D2SF invested in TraceWeave, a startup developing technology to run AI on devices such as phones and cars without connecting to external servers.
**AI take:** The investment signals interest in on-device AI, not a product launch.
### Autonomous A2Z passes 100 permitted vehicles and 1.15 million km
Asia Economy reports Autonomous A2Z has exceeded 100 vehicles with temporary autonomous-operation permits and logged 1.15 million kilometers.
**AI take:** Permit counts and mileage offer deployment indicators, though they need to be read alongside safety records.
### Chainlabs launches its generative-AI search-optimization service
DigitalToday reports that Chainlabs launched HyperGEO, a generative-AI search-optimization service.
**AI take:** Real traffic data will be needed to assess the service.
### Klique expands NVIDIA integrations for enterprise AI resource management
Klique announced new NVIDIA integrations for its enterprise AI resource-management, governance and cost-visibility platform; performance benefits are company claims.
**AI take:** GPU-management tools are extending into governance and cost control, while customers still need to measure actual gains.
### LG shows its Multi V 7 commercial air conditioner with dual AI
LG unveiled the Multi V 7 commercial system air conditioner at HARFKO 2026, combining on-device and cloud AI for operation; performance improvements are LG’s claims.
**AI take:** As AI control enters HVAC equipment, energy savings will need to be tested in real buildings over time.
### PLAUD unveils Plaud One, a wearable for voice-controlled AI agents
PLAUD announced Plaud One, a wearable device for voice commands to AI agents. The company says it has an eSIM charging case, plans a limited release later this year and lists a ¥44,000 direct price.
**AI take:** Voice interaction and independent connectivity could broaden the device’s use, while price and limited supply may shape adoption.
### PredictWind adds an AI Weather Router to its sailing tools
PredictWind launched AI Weather Router, a feature that turns weather-routing and departure-planning results into briefings for sailors and powerboaters.
**AI take:** AI summaries may make complex weather data easier to use, but navigation decisions still require judgment.
### YYForce outlines Workforce OS plans for human and embodied-AI teams
YYForce announced plans to develop Workforce OS, a robotics architecture intended to integrate human workers with embodied AI and technologies including NVIDIA and Unitree; it remains a development plan.
**AI take:** Commercial robotics systems are increasingly framed around working alongside staff, but deployment results remain to be seen.
### Zigbang launches Soma smart-home brand and a child-care door lock
Zigbang reorganized its smart-home business under the Soma brand and launched the Soma Care P1 connected door lock, which can notify parents when a child arrives home.
**AI take:** Home-security devices are adding contextual alerts, making privacy settings and notification reliability important to trust.
### Kakao open-sources two Kanana-2 vision encoders
Kakao released two Kanana-2 vision encoders as open source on Hugging Face, focused on high-resolution images and document and text recognition.
**AI take:** The open encoders give developers new components for image and document-understanding applications.
### AI-focused developer platform Blacksmith raises $66 million Series C
TechCrunch reports Blacksmith raised $66 million in a Series C led by Sequoia at a $1 billion valuation; its platform helps build and test software, including AI-written code.
**AI take:** Growth in AI coding is also drawing investment toward the build-and-test infrastructure around software development.
### Shengshu starts a 15-day livestream of Motubrain 2.0 training
Shengshu Technology began a public livestream on October 21 showing the Motubrain 2.0 training process, with the series planned to run for half a month.
**AI take:** The stream offers a window into embodied-model development, though it is not an independent evaluation.
### Databricks makes its AI Runtime generally available for GPU training
Databricks made AI Runtime generally available as a managed, serverless GPU platform in its Lakehouse for training and fine-tuning deep-learning models on enterprise data.
**AI take:** A serverless GPU service reduces infrastructure management, while customers still need to assess cost and workload fit.
### Krafton releases a preview of its self-developed 30B-class Raon-1 Mini
Krafton released a preview of its self-developed 30-billion-parameter-class language model Raon-1 Mini; the source describes a preview, not a full general release.
**AI take:** A game company entering foundation-model development will need to demonstrate differentiated capability and clear access terms.
### Meta plans a new share class to fund AI and other priorities
The Verge reports Meta plans a new class of non-voting shares for AI initiatives, potential acquisitions and employee compensation, with trading expected to begin in 2027.
**AI take:** The plan is not yet implemented, but signals Meta is preparing another capital instrument for long-term AI spending.
### Recursive releases Flint, a free coding model designed for laptops
TechCrunch reports that startup Recursive released its first model, Flint, a free coding model designed to run on a laptop; the claim that AI helped build it is attributed to the company.
**AI take:** Local execution reduces dependence on cloud APIs, while developers still need to test quality and hardware requirements.
## Rumors
