Google DeepMind's AI agent AlphaProof Code formally verified core security properties of the Signal messaging protocol at the machine-code level. Formal verification, which uses mathematics to prove that code behaves as specified, has been costly and used only in limited areas until now.
Tether AI Research released BotPsy, a 0.8-billion-parameter model built for tool use and agent tasks. It is designed to run directly on devices such as phones.
Adobe unveiled new features for its generative AI app Firefly at its annual MAX 2026 event in Miami. The update adds many AI agents (assistants that handle tasks on their own) and expands the available AI models to 36.
kubegap is a skill, a bundle of tools for AI agents, built by a developer who runs three small Kubernetes clusters alone. The agent reads cluster state, checks 411 items, and writes a report on the problems that need fixing now. It only reads the clusters and changes nothing.
Adobe is reshaping creative software such as Photoshop and Premiere around AI agents. Users describe the result they want in words, and the AI plans and carries out several editing steps. The structure of the original project files stays intact.
REA is an open-source tool that lets AI coding agents analyze the compressed code of apps. The author used it to study the slash menu and block-moving features of the open-source app AFFiNE and then rebuilt them.
Google released Gemini 4 Carbon, an AI model built for complex, multi-step tasks that can work on their own. The company also unveiled Gemini Robotics 2.5, an AI model for robots that work in real-world settings like factories.
Google has released Gemini 4 Carbon, a new AI model built for demanding work such as coding, cybersecurity and scientific research. It is also tuned for tasks where several AI agents (programs that split up and handle work on their own) cooperate.
OpenAI's official channel shared a video from its developer event DevDay 2026 on building with GPT-Live and image generation. It covers how to add voice to agents, which are AI programs that carry out tasks for users.
This is an official OpenAI DevDay 2026 video. It looks at a case where Codex, a coding agent (an AI program that writes and fixes code on its own), kept builds and tests passing for over 1,000 hours.
This video from OpenAI's DevDay 2026 introduces how to use the Codex harness (the base framework that runs and manages AI agents) through the Agents API (an interface for AI that carries out tasks on its own). According to the description, it covers running agents in your own environment, keeping long-running tasks going, and which infrastructure OpenAI manages.
This is a DevDay 2026 talk from OpenAI's official channel. It covers the shift from simple agent loops, which repeat tasks step by step, to architectures where multiple agents work together.
Google DeepMind unveiled AlphaProof Code, an AI agent system that mathematically proves software works as specified. The company says it used the system to verify the core code of the Signal messenger's encryption protocol in about four weeks.
Amazon Web Services (AWS) has launched the AWS Health Impact Initiative (HII) to help people with limited access to healthcare. The program gives organizations that serve underserved groups AI, agents (AI programs that carry out tasks on their own), and cloud computing (renting computing resources over the internet).
Google has released Gemini 4 Carbon, a top-tier AI model built for complex tasks that run over many steps. It targets software development, cybersecurity, and scientific research, where the AI must repeatedly reason and check its work.
This AI Engineer channel video looks at how coding agents (AI programs that write and fix code on their own) have grown. The speaker is Ara Khan from Cline.
Amazon AWS Machine Learning blog · Nov 11, 08:34 KST
AWS released a web search API that lets AI agents fetch current web results in a single call. The post shows how to use it with LangGraph, a framework for running multi-step tasks, to build a daily watch for security risks in open source packages and container base images.
This video from the AI Engineer channel covers how AI agents can make decisions without calling a large language model (LLM) every time. The speaker is Raymond Weitekamp of OpenProse.
OpenAI has added a business hub to ChatGPT for companies. It lets businesses create company profiles and custom agents, which are AI programs that carry out tasks on a user's behalf.