Microsoft has released version 1.25.0 of the Python edition of Agent Framework, its tool for building AI agents. This version adds new features but also includes several major changes that alter how existing code works. Developers who upgrade should review these changes first.
Entire has released a preview of a distributed Git network that copies existing GitHub repositories into regional mirrors. The code stays where it is, and requests from AI agents to download code are routed to the nearest mirror. The US, EU, and Australia regions are currently running, and access is available through a waitlist.
Meta has released Spaces, a custom workspace for use with its AI agent Muse. Users describe the feature they want in text, and Muse builds a workspace suited to the purpose, such as a travel planner or a shopping management tool. It can be used in every region where Muse is available.
This introduces a Korean-language rewriting agent that removes sentences that are hard to follow from reports or blog posts written with AI, then rewrites them to match your own writing style. Rewriting (윤문) means smoothing out sentences while keeping the meaning of the text. The video shows the process of polishing a blog post written by AI.
Meta has unveiled a specialized model trained with its top model, Muse Spark. The model will be applied to the personal AI agent Muse and then to the AI assistant Meta AI. The goal is to move the abilities of a high-performance model into a model that more people can use.
Reports say Meta's new AI agent 'Muse' could collect even bank information. Futurism, a science and technology outlet, reported that Alex Stamos, Meta's former security chief, strongly criticized the AI's indiscriminate data collection.
LangChain has released version 0.10 of its Managed Deep Agents. The update lets agents operate directly on websites that have no API. It also adds subagent features that handle long tasks in the background.
This two-minute video explains how to use AI agents, which are software programs that handle work on their own. It says agents are built from a job description, a single page document, rather than from complicated software.
Germany's Siemens, which makes industrial products and automation equipment, has released a video showing how it uses NVIDIA AI tools to speed up production planning and quality control. It covers using AI agents, programs that divide up and handle tasks on their own, together with digital twins, which are virtual copies of real equipment, to help run factories.
OpenAI has released BioBasicsBench, a benchmark that tests whether AI agents correctly interpret raw data obtained in laboratories. It consists of 100 tasks created with Ligo, an enzyme design company. The top-performing model scored 66.7%.
GitHub says more than 1.2 billion commits were pushed in one week in October, up over 400% from a year earlier. Writing more code does not necessarily speed up the whole job, and this guide explains how to use AI agents (AI that handles several steps of a task on its own) in planning, development, verification, and operation.
Meta has launched Spaces, a new feature for working with its AI agent Muse. Spaces are separate conversation areas organized by topic. The report is based on coverage by TechCrunch.
Anthropic tested Claude agents in the game EVE Online, checking their trading skills and their ability to spot sabotage by teammates. The agents made profits in a simulated market, but they rarely caught saboteurs on their own team.
AWS Agent Registry is now generally available in the AWS GovCloud (US-West) Region, which public sector customers commonly use. Agent Registry is a service that gathers AI tools and agents in one place so they are easy to find and manage.
A guest essay published by Anthropic argues that AI agents should not offer just one most plausible scientific model. Instead, they should show several possible models along with their weights, because multiple explanations can account for the same measurements. Based on how researchers work, the essay proposes this standard.
OpenAI has teamed up with Hacktron, a security research company, to look for weaknesses at the points where multiple systems connect. The partnership focuses on how AI agents link to services. It also expands access to OpenAI's advanced cybersecurity models.
Amazon AWS Machine Learning blog · Nov 7, 02:13 KST
The AWS Machine Learning Blog introduced a server that helps AI agents (AI programs that do tasks on behalf of people) access a company's data lake. The server lets multiple agents find and use data through the same channel. It was reportedly built for Amazon's fulfillment center planning team.
Amazon AWS Machine Learning blog · Nov 7, 02:09 KST
This part covers how to expand the work agent built in Part 1 into an environment where multiple agents and tools are used. A gateway at the entrance receives all requests and first filters them using policies. A second agent running on another platform is connected in the same way.