Perplexity added a voice mode to its AI team members on iOS and Android, so Max subscribers can start tasks by phone, check progress, and change direction while the work is running. The feature extends the company's Perplexity Computer agents to mobile use.
A Python package for Microsoft's AI agent development tool, Agent Framework, was distributed with code that steals cloud account credentials. The malicious code reportedly runs every time Python starts.
This is part 6 of a series on using AI for practical work, covering Skills (part 2). It follows part 5, which explained that AI agents need context, fixed code for repeated tasks, and prompts to guide work procedures.
Figma now lets users give instructions to its AI agent directly from dev mode, the view developers use to inspect designs. The change lets developers prompt the Figma agent without leaving that view.
OpenAI Developer changelog (mirror) · Nov 19, 09:00 KST
OpenAI published a guide on making GPT-Live voice agents sound natural for specific languages and regions. It covers picking a built-in voice, designing a synthetic voice from a text description, and using a recording of a consenting speaker. The guide also includes example prompts for Indian English, Spanish from Spain, and Emirati Arabic.
ElevenLabs has released Scribe v3 and Scribe v3 Live, two speech-to-text models for turning recorded and live speech into text. Both are available through its developer API, and Scribe v3 Live also works inside its ElevenAgents platform for businesses.
Microsoft CEO Satya Nadella said AI agents could reshape how the internet makes money, especially for businesses built on ads and human visitors. He spoke at the New York Times DealBook Summit and compared the change to the earlier move from web portals to search engines. He said companies that refuse AI agents may lose out to rivals that accept them.
Amazon AWS Machine Learning blog · Nov 19, 06:48 KST
Airbnb runs its internal coding agents, AirChat and AirChat Remote, through an LLM gateway on Amazon Bedrock. Thousands of engineers use them weekly, and about two thirds of their requests recently went to newly released models. Airbnb tests each model on its own tasks before rollout and then checks production data against those results.
OpenAI showed how AI agents can handle routine paperwork in US healthcare, such as insurance pre-approvals and medical records. In its tests, the agent drafted documents while doctors made the final decisions. OpenAI said the tools can cut staff workload and speed up patient care steps.
AWS has made Amazon OpenSearch Memory generally available. It is a serverless feature that gives AI agents long-term memory, so they can recall past decisions, earlier conversations, and user preferences without replaying the whole chat history.
At Microsoft Ignite 2026, Microsoft urged its more than 500,000 partners to expand into AI agent operations, security, and cost management. Its "Customer Zero" approach asks partners to use AI in their own work first and then sell what they proved to customers.
LangChain says many teams' coding agent bills are climbing fast because no one tracks the spend. A single feature can use several AI coding tools, and each logs usage differently, so teams can't see what each one costs or delivers. LangChain proposes a shared view across tools, built on its LangSmith platform.
Perplexity says its Brain personalization system for the Computer agent can now create and update reusable agent skills from users' past work. It reports over 80,000 personal skills written for more than 20,000 users between August 1 and November 14, 2026. The feature is on by default for US Max subscribers and can be turned off.
Internal chats show Meta staff raised alarms about its Muse AI agent. Staff said it seemed to read private encrypted messages, mix up different users' conversations, and show unrelated sensitive info like bank balances. Meta called these isolated incidents caused by hallucinations, meaning the AI made up false information.
Amazon AWS Machine Learning blog · Nov 19, 04:05 KST
AWS published a guide for connecting Amazon Bedrock AgentCore Gateway to private APIs and MCP servers in a network where all traffic must pass a central firewall. The network team keeps control of every subnet, and the Gateway's requests and replies go through the inspection VPC in both directions. The setup places a VPC Lattice resource gateway in the network account and routes it through a transit gateway. The guide was tested in a lab in the us-east-1 region.
Amazon Science describes how AI agents can write mathematical proofs that a computer checks step by step, so made-up claims get caught. The approach connects an AI agent to a proof assistant, a tool for writing and verifying formal proofs, through the Model Context Protocol (MCP, a standard for linking AI agents to tools). The article focuses on Isabelle and shows AI-written proofs of classic results, such as the square root of two being irrational.
Amazon AWS Machine Learning blog · Nov 19, 03:54 KST
AWS held an inaugural AWS Quick User Group meetup in Arlington, Virginia, on October 7, 2026. Data leaders and developers talked about how to prepare company data so AI agents can use it. One firm, Financial Independence Group (FIG), shared how it is using AWS Quick for its analytics.
Microsoft AI has released seven new in-house models for coding, security, images, speech, and routing, all at once at its Ignite conference. The company says they match rival models at a much lower cost, and it is aiming them at AI agents that run around the clock.
Scientific computing is entering the era of agentic AI. Bringing simulation, AI, and agents together gives researchers new ways to tackle complex problems and accelerate scientific discovery. Watch Ian Buck, VP of Hyperscale and HPC at NVIDIA, highlight the latest cutting-edge breakthroughs revolutionizing the scientific computing community. https://blogs.nvidia.com/blog/ian-buck-sc26/ https://blogs.nvidia.com/blog/sc26-wrapup