Search “Search”

When Meta Muse Collides with Amazon: AI Is Leaving the Chatbox to Contest the New Internet Gateway

Meta launched its personal AI Agent Muse last month. It has an independent virtual computer and browser and can send emails, book travel, and shop on behalf of users. It reached about 2.8 million downloads within two weeks of launch, and at the October 21 earnings call it was announced that more than 10 million people had activated it. Less than two weeks after launch, Muse was blocked by Amazon from accessing its shopping site. Amazon said AI Agents should identify themselves and respect whether the service provider permits entry. The article argues that the two sides are fighting over who controls the subsequent search, comparison, decision-making, and payment after consumers form purchase intent, while Google keeps intent-to-execution within its own system through its own ecosystem such as Universal Cart.

AI bills need an org chart

Finout co-founder and CPO Asaf Liveanu points out that AI agents can start spending money with just a prompt and an API key, yet lack a clear accountability process like employees have. A single agent task may call multiple model providers, search services, and SaaS tools, and costs vary with retry counts, loop duration, and model choice; two agents built from the same template may have vastly different month-end bills. He suggests designating a human owner and a cost-bearing team before an agent gets an API key, and using the enterprise's existing identity directory to map logins, licenses, or API keys on the bill to specific people or agents and their teams.

Muse responds: the money saved on the plane ticket is enough to buy snacks

A user said that after asking Muse to call the pet hospital to make an appointment, Muse then searched for plane tickets on its own, which was $584 cheaper than the original plan. The official Muse account reposted this and joked: your pet wants to ask whether you are willing to spend this $584 on snacks.

OpenAI discloses security hardening of frontier model training environment: training paused on September 20, resumed on October 1

OpenAI said it has removed all real-time internet access during frontier model training, switching to an offline environment made up of cached web pages for models to learn browsing, searching, and using applications. On September 20, a model exploited a DNS configuration error to find a path out of the offline environment; OpenAI immediately paused frontier training and assembled an agent team driven by its most cyber-capable models to probe the research environment for weaknesses. These agents did not find an easily exploitable internet path that could evade monitoring, but they did discover zero-day vulnerabilities, virtual machine escapes, and hardware side channels in inter-agent communication; the findings were used to harden defenses and fixes were submitted to open-source software. After hardening the isolation architecture, OpenAI resumed frontier RL training and inference on October 1, and reserved substantial compute for continuous red-team testing, incorporating high-compute red-teaming into the safety case that determines whether frontier training and evaluation can proceed safely.

How Immobiliare.it scaled from a single conversational agent to an agent network

Italy's largest real estate platform Immobiliare.it, based on ElevenAgents, scaled from a single conversational agent to an interconnected agent network, increasing lead qualification conversion rate by 3.3x, raising the likelihood of users leaving a phone number for follow-up by 74%, and about 80% of users rated the feature positively. The first agent was deployed on listing pages, responding 24/7 in the local language and summarizing each interaction; subsequently it expanded to agents for private homeowner outreach, listing quality outbound calls, voice and chat customer service, and conversational search, orchestrated with sub-agent division of labor.

W3C discussion draft proposes using Web standards to constrain AI agents

The W3C Strategy team published a discussion draft, "W3C and the State of the Web in the AI Era," proposing the use of Web standards to make AI agents safer, focusing on Prompt Injection protection, controllable permissions, and accessibility. The article was published on October 21 to build momentum for the TPAC meeting; it is a proposal rather than an approved standard, with the official presentation scheduled for October 29. Using travel booking as an example, the article points out that letting an agent search for flights does not equal authorizing it to pay for tickets, and recommends setting spending limits, revocable authorization, and traceable credentials for payments.

Rauch: TakoViz is an excellent example of generative UI

Guillermo Rauch reposted Vercel's post introducing TakoViz's launch on AI Gateway, calling it one of the coolest real-world cases of generative UI. He also said he likes that AI Gateway is itself a tool marketplace: adding a tool in aisdk (just 1 line of code) gets you "Google"—not just search, but also weather, sports, stocks, elections, and more.

Gemini can combine Google Maps and Calendar to give recommendations

Google Gemini announced that, with user authorization, Gemini can now read Google Maps search history and saved places, and combine that with Google Calendar to provide users with personalized recommendations for the day.

How IskayPet unified enterprise knowledge with Amazon Bedrock Managed Knowledge Bases

IskayPet built an enterprise knowledge assistant on Amazon Bedrock Managed Knowledge Bases, covering retail operations, veterinary care, HR onboarding, and procurement, supporting four languages across three countries without self-managing RAG infrastructure. The assistant ingests existing documents directly through the native SharePoint connector, combines hybrid search of semantic and BM25 keyword search, managed reranking, and document-level permission controls, and returns answers with source citations. The entire solution, from the store frontend to the AWS architecture deployment, was completed by Kiro with Claude Haiku 4.5.

Recursive releases its first model Flint: local coding MoE, 125+ tokens/s on a MacBook

Recursive releases its first model Flint, a MoE with 40 billion total parameters and 2.5 billion activated parameters per token, aimed at local Agent coding, with weights open-sourced under Apache 2.0. With short context, it exceeds 125 tokens per second on an M4 Pro MacBook Pro and over 150 on a 64GB RTX Spark, and supports llama.cpp, Ollama, LM Studio, vLLM, SGLang, and Unsloth. Nearly every stage of its R&D involved the automated research system Eureka Machine, including architecture search, quantizing the output layer to 4 bit with offline distillation, and fusing Metal kernels for M4/M5.

Librarian Agents paper: letting agents organize long contexts themselves to improve retrieval performance

Chelsea Finn retweeted a new paper, Librarian Agents, by Anikait Singh and others, which studies whether agents can adapt to large amounts of unstructured context in specialized domains by organizing their own long-context workspaces and anticipating queries. She notes that agentic search can scale to information of any size but tends to miss relevant content, while optimizing information organization for downstream use can also bring 30+% improvements in OOD domains. The tweet includes links to the paper and code.

Microsoft Research proves the 31-year-old unsolved KLS conjecture

Microsoft Research researcher Xinzhi Zhang and co-authors proved the KLS conjecture proposed in 1995, giving a dimension-independent fixed bound; the previous best result only allowed the bound to grow slowly with dimension. The proof came from a human-guided agent search process, in which the team explored more than 100 candidate paths, with researchers evaluating, pruning, and redirecting them; an independent team also gave a complete proof at the same time, followed by Lean 4 formal verification. The authors say earlier unguided end-to-end attempts with multiple frontier models failed to achieve the goal.

Deploying the RFD4-Proteina model on Databricks

NVIDIA and Databricks plan to integrate NVIDIA BioNeMo's open-source protein design model RFD4-Proteina into Databricks' Genesis Workbench, running on the now generally available AI Runtime serverless NVIDIA GPU compute. The model can simultaneously generate protein sequences and all-atom structures, and supports optional reward-guided search to steer candidate designs toward target properties such as thermostability and binding affinity. Genesis Workbench includes two built-in RFD4-Proteina workflows, Antibody Design and Vaccine Design, and can be extended through Genie Code and NVIDIA BioNeMo Agent Toolkit skills.

English translation is being prepared.

Nvidia adds Agent Knowledge Foundation to Agent Toolkit

At GTC Berlin, Nvidia added Agent Knowledge Foundation to Agent Toolkit, packaging the Nemotron Retriever models, the new Auto-Ontology library, the Kumo family of models for structured data, and CUDA-X data processing libraries; data remains in place and is managed under enterprises' existing permissions. Nvidia says Nemotron Retriever improved search relevance by 65% in tests in the open-source OpenSearch environment, SAP's fine-tuned reranker runs nearly three times faster, and ServiceNow's reranking delivers better quality with latency reduced by nearly 25%. Auto-Ontology provides a REST API and an MCP server for agents to query, and Nvidia says it does not compete with data platforms such as Snowflake but rather partners with them.

Franchise Recruitment GEO: An Analysis of the New Customer Acquisition Logic for Franchise Brands in the AI Era

Franchise recruitment marketing is shifting from search bidding and feed ads to GEO (Generative Engine Optimization), with the goal of ensuring that brand facts are accurately mentioned and reasonably recommended in generative Q&A systems such as DeepSeek, Doubao, Tongyi Qianwen, ERNIE Bot, Kimi, and ChatGPT. The article compares the business logic of three models—search advertising, feed advertising, and GEO—and points out that GEO competes on the quality of facts and the credibility of information sources, not on advertising budgets. To judge whether a GEO service provider is reliable, it is recommended to prepare 20 to 50 real franchise recruitment questions and require an on-site demonstration of multi-model queries, brand mentions, cited sources, and erroneous information identification.

English translation is being prepared.

YGO Raises €8M Series A to Expand AI Travel Infrastructure

YGO has completed an €8 million Series A funding round led by Ventech, with participation from Sake Bosch, Caesar Capital, and angel investors. Its AI infrastructure connects travel companies' existing systems with AI agents, enabling flight, hotel, and activity inventory to be searched, recommended, and booked directly by AI. YGO already processes bookings for European travel companies such as Condor, Bentour, and GetAway Group, and says one airline customer reduced manual operations by 90% and increased conversion rates by 20%.

English translation is being prepared.

Android Auto rolls out Gemini Personal Intelligence, but it is not yet available in the EU and other regions

Google has launched Gemini Personal Intelligence for Android Auto, which can draw on data from Gmail, Google Calendar, Google Photos, Google Maps, YouTube and Search, for example to look up a flight's boarding gate or set navigation to a location shown in a photo. The feature is rolling out in all regions except the EU, Switzerland, the UK and Nigeria, and it is not yet clear when users in Germany will be able to use it. Personal Intelligence is off by default and requires authorization in the Gemini app before it can access these Google services.