AI Moves Deeper into Search, Teams, and Devices

AI Moves Deeper into Search, Teams, and Devices

Key Takeaways

  • AI is becoming more deeply embedded in existing systems: Google, Microsoft, and Apple are betting on persistent search, team agents, and proprietary platforms.
  • Capabilities are advancing unevenly: METR’s updated analysis shows progress on longer tasks, but weaker performance on tasks requiring sophisticated judgment.
  • Infrastructure and control are becoming factors in location decisions: Debates over data centers, financing, and access rights show that AI is not just a model question, but an infrastructure question too.
  • Robotics is moving closer to real-world operations: New motion control technology and coordinated cleaning robots target practical applications, while researchers study trust and flexible designs.

🔬 Models & Research

AI’s most misunderstood chart just got a major update
November 20, 2026

METR has updated its much-discussed time-horizon plot, which assesses how long the tasks are that models can complete autonomously. The results point to continued progress, but also show that coding performance is not the same as judgment on complex tasks.

Anthropic mathematician says he used Claude to solve a prestigious Millennium Prize Problem
November 20, 2026

Anthropic mathematician Levent Alpöge says he proved the Riemann Hypothesis with the help of Claude; independent verification is still pending. If the proof holds up, it would be a remarkable test of whether language models can go beyond assistance in open-ended mathematical research.

Model overload: Which AI do I need? | Rock. Paper. AI.
November 20, 2026

The podcast discusses how to compare AI models for specific tasks rather than judging them solely by general benchmarks. For teams, this means that their own tests using real workflows may be more meaningful than frequent model changes prompted by new rankings.

💼 Business & Markets

Report: Nvidia close to acquiring AI startup Inception for more than $500 million
November 20, 2026

Nvidia is reportedly close to acquiring Inception, which develops language models using diffusion techniques. The potential acquisition shows that Nvidia wants to integrate alternative model architectures and the teams behind them into its AI ecosystem, in addition to chips.

Fallout from the Nvidia scandal involving Jensen Huang continues
November 20, 2026

A TechCrunch podcast examines Nvidia’s latest business results, the dispute over corporate leadership, and the halt to H200 sales in China. For the AI market, the key issue remains how export restrictions and governance will affect demand for its essential accelerators.

Studies show significant AI-driven job losses in China and warn that 70 million jobs are at risk
November 20, 2026

A Bloomberg analysis of nearly ten million Chinese job postings suggests that AI is putting further pressure on an already weak labor market, especially for entry-level workers and creatives. Companies should therefore weigh productivity gains against changing entry-level career paths and skill requirements.

Chips can withstand the conditions in space—but what about their financing?
November 20, 2026

Building data centers in space faces not only technical hurdles, but also the question of how expensive chips can be financed without the usual collateral. Since loans on Earth are often secured by the resale value of GPUs, a lack of liquidity could further hinder the business model.

This 16-year-old founded an AI missile startup. His goal: “cost per kill”
November 20, 2026

16-year-old Henrijs Dandzbergs founded defense startup Polemos, which develops AI-powered missiles and puts the cost per target at the center of its pitch. The case shows how quickly autonomous weapons technology is entering the startup economy—and how urgently procurement, safety, and oversight need to keep pace.

⚖️ Policy & Regulation

Trump threatens “massive taxes” if AI data centers are built outside the US
November 20, 2026

Donald Trump is threatening companies with steep levies if they build AI data centers outside the US. Together with local restrictions on new facilities, this is increasing uncertainty around site selection, energy planning, and long-term infrastructure investment.

🛠️ Tools & Products

Google rolls out a collapsible sidebar for AI searches on its desktop homepage worldwide
November 20, 2026

Google is rolling out a sidebar worldwide that lets users revisit previous AI searches, search through threads, and start new conversations. The always-available interface makes AI Mode more like an ongoing workspace than a one-off search feature.

Signal blocks Gemini and Siri from reading the screen
November 20, 2026

Signal is restricting screenshots on iOS and Android to prevent built-in AI agents from capturing messages and other on-screen content. The move shows that privacy will increasingly depend on how much control operating systems give apps over agents’ access.

Apple is developing its own AI agent and making it harder for competitors to access its devices
November 20, 2026

According to a job posting, Apple is working on its own agent, with an emphasis on privacy, permission systems, and sandboxing. At the same time, tighter control over device access could hinder competing agents and strengthen Apple’s platform advantage in everyday AI use.

Hermes: Popular open-source agent comes to Microsoft Teams
November 20, 2026

Nous Research is bringing its open-source agent Hermes to Microsoft Teams, where employees can delegate tasks such as analyses, document drafts, and inbox management. Running it in isolated Azure container sandboxes combines workflow automation with a clearly separated execution environment.

Six AI-powered gadgets that could make your home office smarter
November 20, 2026

The roundup features six AI devices for the home workspace, including a pen that combines handwritten notes with audio recordings, transcription, and summaries. Such specialized devices can reduce routine work around meetings, but should be evaluated based on how they process data and how much time they actually save.

🤖 Robotics

Dongwoon Anatech enters the robot motion-control market, powering the future of physical AI
November 20, 2026

South Korean chipmaker Dongwoon Anatech is entering the market for humanoid robots and autonomous vehicles, drawing on its expertise in precise motor control. An additional supplier could diversify supply chains and make motion control—a key requirement for physical AI—a more competitive field.

KEENON Robotics showcases multi-robot collaboration and explores humanoid-assisted cleaning
November 20, 2026

At ISSA 2026, KEENON Robotics presented collaboration between multiple robots and approaches to cleaning with humanoid assistance. For operators, the question is whether coordinated systems can handle complex building tasks with less manual coordination and more reliable operation.

User-initiated and system-initiated human feedback in human-robot collaboration: effects on trust in robots
November 20, 2026

A study involving 48 participants compared whether requested or spontaneously initiated feedback affects short-term trust in a robot. The findings highlight how feedback-loop design can shape collaboration and perceived safety—even when the robot’s behavior does not change.

A virtual central ball joint for reconfigurable truss robotic systems
November 20, 2026

Researchers present a passive ball joint for truss robots that allows large three-dimensional movements and rapid reconfiguration. The mechanically simple design could make modular robotic systems more flexible without abandoning their underlying structural concept.