Post 5 of 8Live
AI agents: news from primary sources
A live digest: releases of Claude Code, Codex, Cursor, the MCP specification and agent frameworks. Collected from official blogs, changelogs and GitHub releases, every note links to the primary source.
Alexander Mihalkevich · Fact-checked October 5, 2026
Agent tools change every week: new permission modes, configuration formats, protocol versions. This page is a live digest. Below are the latest notes on releases and changes in agent tooling. They are collected by our news feed: it checks primary sources every 15 minutes and writes a short note with a link to the original.
What goes into the digest
Notes from the “Agents” and “Developer tools” topics:
- Claude Code and Codex CLI releases on GitHub (patch releases are filtered out);
- the Cursor changelog;
- the Model Context Protocol blog and MCP specification releases;
- blogs and releases of agent frameworks, and changes to AI tools from GitHub and other developer platforms.
How to read agent news
A piece of tool news matters not in itself but for what it changes in your work. Before you upgrade or rebuild your process:
- Open the primary source. Check the date, the version number and the exact wording. Retellings lose the caveats.
- Tell the statuses apart. “Research preview,” “beta,” “experimental, off by default” and “generally available” are different things. For example, agent teams in Claude Code are marked experimental and are off by default.
- Look for changes to default behavior. The most important news is what changes the starting permission mode, the default model, limits or the configuration format. It is easy to miss, and the consequences show up immediately.
- Test it on your own task. Run the updated agent through your task set before changing the process for the whole team. How to build such a set is covered in the first post.
Where to follow it yourself
If you want to get changes directly, subscribe to repository releases on GitHub (Watch → Custom → Releases) and to the official changelogs. Our feed collects the same things in one place and writes a short summary in Russian.
Latest news
- New agent skill: Amazon SageMaker optimized generative AI inference for your coding agent
Amazon SageMaker optimized generative AI inference is available via the Agent Toolkit, giving agents inference optimization skills.
AWS Machine Learning · October 5
- Agentic retrieval with LangChain and Amazon Bedrock Knowledge Bases
Amazon Bedrock Knowledge Base with LangChain enables building RAG applications with agentic retrieval for multi-part questions.
AWS Machine Learning · October 5
- The Agent Said It Was Done. The Database Disagreed.
GitHub Copilot now supports APIs for code review requests and setting effort levels, with 'Balanced' as the default.
Hugging Face · October 3
- How Rogo ships agent-written code to production in 5 minutes on Vercel
Rogo uses Vercel to ship agent-written code to production in 5 minutes, with over 73,000 deployments per month.
Vercel · October 2
- Build Applications on NVIDIA BlueField Faster with NVIDIA DOCA Agent Skills
NVIDIA DOCA Agent Skills accelerate the development of AI agents for specialized infrastructure software like NVIDIA BlueField.
NVIDIA Developer · October 1
- Vercel Agent now installs private packages from npm and custom registries
Vercel Agent can now install private dependencies from npm and custom registries using credentials stored as shared environment variables.
Vercel · September 30
- Let skills in Gemini tackle your most repetitive tasks
Gemini now features skills, replacing 'gems', to automate repetitive tasks with reusable custom instructions.
Google Gemini · September 30
- Tracing Agent Harness Behavior with NVIDIA NeMo Relay
NVIDIA NeMo Relay helps trace agent behavior, identifying inefficient task paths, failed searches, or file read errors.
NVIDIA Developer · September 30
Terms in this post