#agents

Meta has released Muse from Meta, an official Chrome extension. Instead of only using a remote browser on Meta's side, Muse now sits in the side panel and works in your own Chrome: filling in forms, comparing prices and chasing refunds on sites you're already logged into, and asking before it spends money. What it can see, how to start, how to stop it anytime, and the safe setup we recommend.

Google has released Gemini 4 Hydrogen to replace Gemini 3.8 Flash. Its overall score is slightly above Argon, it runs about 2.5 times faster, and it has a free tier in AI Studio and the Gemini API, with paid use at $1.50 per million input tokens. We cover the benchmarks, pricing, where to use it, how to migrate from older models, and a prompt you can give your AI agent to switch models for you.

A written customer profile and a prospect list. Companies are filtered first, then people. Every email has a source. You spot-check the list, then send to a small batch.

Give it a website and get back a teardown of how the company acquires customers. Every judgment is tied to evidence, and anything that could not be checked is listed separately. Then turn it into your own 30-day plan.

Google's Gemini Decisions API and the new Gemini Decision 1 model handle closed questions: yes or no, pick one, or score on a scale. Up to 1000 questions per request, about 0.2 seconds per response, and free output. Useful for choosing tools for an AI agent, routing requests, and running safety checks. Includes a prompt you can hand to your AI agent to try it.

Meta's smallest and cheapest Muse model, Muse Ember, is built for repetitive work: tool calling, small coding tasks, routing, and extraction. It costs $0.10 per million input tokens and offers a 1M-token context window. Covers pricing and specs, how it compares with Claude Haiku 5.5 and GPT-6 Luna, and where it fits in an AI agent's workflow.

Beginner tutorial: Download the Nod desktop app, connect a ChatGPT subscription, a free Gemini key, or a local model, link only the email and calendar you need, set clear limits on each task, and approve anything that sends or pays. Includes 2 copy-ready prompts.

Gemini 4 Carbon is now open to everyone through the paid API and Google AI Ultra. It's built for long-running agent work and cybersecurity. Here's how it splits work with Argon, when the price doubles, and what Antigravity's /swarm does.

ERNIE 5.5 Flash Preview is live on Baidu's Qianfan platform and OpenRouter, with open weights promised by November 25. What it's good at, where it lags on benchmarks, and how to try it for free.

Google's A2UI 1.0 lets an agent say what belongs on screen while your app paints it with its own native parts: forms, charts and confirmation cards. Jira and Alation already use it, and it runs in Claude and Gemini through MCP Apps. Includes a prompt that walks your AI agent through the quickstart.

Hermes Agent v0.23.0 adds a web UI (Webapp): the full Desktop app workspace in your browser, served by your own gateway. Sessions, group chats, plugins and files sync with the Desktop app. Off by default, three commands to turn on. Set up sign-in before opening it to the local network or internet. Includes a prompt to hand to your AI agent.

Cursor launched Voice Chat on November 6: live voice calls with an Agent in the IDE, Agents window, web, and iPhone. Dictate tasks to the coding Agent and hear brief progress updates while it works. On iPhone, say "Hey Siri, call Cursor" to start a call. Powered by Grok Voice; no data is retained in Privacy Mode.

Intermediate tutorial: install Polymarket's official command-line tool. The API is free and needs no wallet. Have Claude Code or another agent check markets, look at odds trends, search the web for information, give its own probability and compare it with the market, log predictions in a table, and score them with the Brier score after they settle. If you want to trade, use your own account and place orders yourself.

Hand the tutorial to an Agent and it works through it to the end; you only confirm key steps such as logging in and paying, so learning and putting it into practice happen in one go.

Use when the user wants answers to AI agent learning questions that come from CWYZ course content and include sources. Calls the CWYZ Q&A API with an API Key: it first classifies the question, then finds answers in the matching site sections and returns source links. Free accounts get 1 question per day, members get 30 per day. Use case: Let your agent ask CWYZ questions directly and get answers based on course content, with source links.

Tutorial: Add a Gmail Connector to Meta's Muse, set its permissions to read-only, then use one sentence to schedule daily or weekly tasks that have it automatically read new emails and send a summary report sorting them into three groups: needs a reply, bills, and can be ignored. Steps and screenshots come from Meta's official Help Center.

Beginner tutorial: you can write a Skill without coding. Step 1 is finding your repetitive work, step 2 is running it through manually first, step 3 is explaining the requirements to AI so it writes them into SKILL.md, step 4 is running and refining it, and step 5 is sharing it with others to use.

Hermes Agent, an open-source project from Nous Research, is the same kind of product as OpenClaw but works quite differently: it creates skills on its own from conversations, improves them as it's used, remembers you across sessions, and can dispatch sub-agents to work in parallel. In our testing it ran alongside OpenClaw and could revive it when something broke. MIT licensed.

Intermediate tutorial: Don't have the Agent run scheduled tasks by hand every day. Have it write the task as one Python script, hand it to the system's built-in cron or Task Scheduler, and run it at 0 Tokens per run. When the script finishes, it sends you a message saying whether it succeeded or failed.

Zoho International has a CLI, so an Agent can easily take over sending and receiving mail. Zoho China has no CLI and only a limited API. The workaround: CC the mail to Gmail, have the Agent read it with gws and write a draft, then you review it and send it to the client yourself.