Brian Douglas of Paper Compute explains how to record the corrections people make during conversations with coding agents (AI tools that write code) and put them to use. The talk was posted on the AI Engineer channel.
This video looks at the problems that appear when AI coding agents use command-line tools (CLIs). Speaker Ayush Bhardwaj analyzed about 20,000 tool calls from his own coding sessions.
This video argues that reading an AI agent's skill file alone cannot show what it will actually do. The speaker is Roey Ben Chaim of Zenity. He explains that even an official diagnostic skill can read cloud credentials and query an instance metadata endpoint, which can look like information-stealing malware.
Microsoft's C++ language server for GitHub Copilot CLI now includes a compile commands extractor for MSBuild and runs it automatically. Agents no longer need a separate setup step to analyze your project accurately.
Meta has extended its personal AI agent Muse to Apple Watch and Google Wear OS smartwatches. Users can text Muse, dictate to it, or make calls to it from the watch.
Amazon AWS Machine Learning blog · Nov 11, 01:59 KST
Tata Consultancy Services (TCS) built a tool that turns requirements into executable tests, using Amazon Bedrock AgentCore (a service for building and running AI agents). The goal is to speed up software testing for banking, financial services, and insurance clients.
eBay seems to have blocked outside AI agents, including Meta's Muse, from browsing its site. A reader's Muse agent had been compiling daily watch listings for weeks, but this weekend it could no longer get eBay details.
Rescale unveiled a framework for customizing engineering AI agents, which are programs that carry out tasks on people's behalf. Kairos Power and ORNL's manufacturing facility are using the platform for heat, flow, and shielding simulations and reports on the 50-megawatt Hermes 2 reactor.
Nvidia has donated OpenShell, an open source tool that helps companies secure and manage AI agents (programs that carry out tasks on their own), to the Cloud Native Computing Foundation. It also released new tools for running AI work on Kubernetes.
Workday built an internal platform for creating and running AI agents, which are programs that carry out work on their own, for HR, finance, and IT. The team used LangGraph to design each agent's workflow and LangSmith to trace and evaluate how the agents behaved.
Google DeepMind, working with Artificial Analysis, released MKA (Multimodal Knowledge Agent), a benchmark that tests how well AI agents (AI that carries out multi-step tasks on its own) handle work where visuals and layout matter, such as blueprints, schematics, reports, and spreadsheets. It includes 200 tasks written by domain experts across 10 fields and 83 occupations.
Google has released its newest AI model, Gemini 4 Carbon, to the general public. It first went to a small group of cybersecurity defenders two weeks ago. Google says it is its strongest model so far.
Herdr CKD is an open-source tool built to reduce the problem of AI coding tools declaring their own work finished. It separates the agent that writes code from the agents that review it, so they check each other.
Meta has released its Muse AI assistant on Apple Watch and Google's Wear OS smartwatches. Users can now talk to Muse from their wrist by voice call, text, or dictation.
CodeRabbit co-founder and CEO Harjot Gill explains how his team uses OpenAI models to understand, review, and check AI-generated code for security. He also covers how they decide whether the code is ready to ship.
NVIDIA introduced Agent Substrate, an open-source runtime for running AI agents at large scale. It saves the state of idle agents and restores them onto free workers, so a small number of servers can handle many agent sessions. The project was donated to the Cloud Native Computing Foundation (CNCF) Sandbox.
Google DeepMind used AI agents to mathematically prove that the encryption behind the Signal messaging app keeps its promised security properties in the code that actually runs. The team says it is the first time a widely used encryption protocol has been verified this deeply with AI. The check took four weeks.
Anthropic's new model Fable 5.2 completed 41.3% of real freelance jobs on the Remote Labor Index, the top score so far. That is more than 16 times the 2.5% best score from a year ago. The test checks whether AI agents (AI that carries out multi-step tasks on its own) finish work well enough for a paying client to accept.
Adobe announced at its MAX event in Miami that it is adding more AI to Photoshop, Premiere, Illustrator, and InDesign. It also gave its Firefly design tool a creative AI agent.