Good Morning! Here's what I have for you in today's newsletter:
Google rolls out Gemini 4 Hydrogen for fast, everyday help
Anthropic brings Claude into your video calls with Claude Meetings
OpenAI lets you set custom instructions and rules for your ChatGPT dot
Build a Claude Code team to write and check your weekly brief
4 new AI tools worth trying today
AI MODELS
Google rolled out Gemini 4 Hydrogen, a model built for fast, everyday tasks in the Gemini app. It’s rolling out globally on desktop, mobile, and web. On the free plan, Gemini selects Hydrogen automatically when it fits the request.
Google says Hydrogen follows instructions about tone, format, and other details more closely. It can also turn product comparisons or dense material into tables and visual summaries.
For larger tasks, it breaks the work into steps and can draw on uploads, Gmail, Google Drive, and connected apps such as Airtable and Dropbox. Google says it’s also better at choosing the right tools and skills.
Hydrogen can build working layouts for interactive prototypes. Google says it takes fewer prompts to get to something you can test and refine.
Hydrogen shows how much competition now happens in the model most people get without picking one. For an everyday assistant, following instructions and choosing the right tool can matter more than winning a difficult benchmark. With Google routing some free requests to Hydrogen automatically, the rollout also tests how well that model choice works when users leave it to the app.
AI TOOLS
Anthropic introduced Claude Meetings, which lets Claude join a video call, answer questions, and build documents during the discussion. The beta is rolling out on Team and Enterprise plans, starting with Google Meet. Other platforms are planned.
Claude appears as a visible participant. Ask a question out loud or in the meeting chat, and it can answer while the conversation is still going.
Ask Claude to start a document, whiteboard, or other artifact. The team can see the draft take shape and correct it before the discussion moves on.
After the call, the notes, action items, transcript, and anything Claude created appear together on the meeting’s page in Claude.
Claude Meetings marks a shift from AI note-taking to real-time collaboration. Agents can now draft plans and documents while decisions are still being made, giving teams a chance to catch misunderstandings before the call ends. As these tools become part of workplace meetings, companies will need to distinguish AI drafts from decisions the team has approved
AI AGENTS
OpenAI updated dots, ChatGPT’s always-on agents, with custom instructions and rules for when to ask permission. The company also says dots reply faster, use less jargon, and follow through more reliably.
Give your dot instructions in its thread, from “write in lowercase” to “show me a draft before sending any email.”
Set rules for actions that need your approval. A dot can keep working in the background and check in when it reaches one of those steps.
On mobile, find the settings under Customize > Behavior. On web and desktop, they’re in the dot’s menu. OpenAI says custom instructions and rules cannot override built-in safety requirements.
Standing rules give always-on agents a clearer division of responsibility. Drafting an email and sending it can be separate decisions, even when the agent works across both steps. As agents take on longer assignments, reliably pausing at the right moment will be central to how much authority people and organizations are willing to give them.
HOW TO AI
Three Claude Code agents research, write, and check your Monday brief. You review it.
Give each Claude Code agent one job: the researcher finds updates, the writer drafts the brief, and the checker reviews it against your rules. You’ll need Claude Code, the Claude desktop app, and a Pro or Max plan.
Step 1: Set up the folder
In Terminal, run claude update, then mkdir ~/weekly-brief, cd ~/weekly-brief, and claude. Confirm that you trust the folder you just created when prompted.
Ask Claude to create a CLAUDE.md with your topic, 5 to 10 trusted sources, your audience, and words to avoid. Set the format to five items under 400 words, each covering an update from the last seven days and linking to its source.
Step 2: Build the researcher
Ask Claude to create a project subagent named researcher in .claude/agents/. Use Sonnet with only WebSearch, WebFetch, Read, and Write. Tell it to find 8 to 10 updates from the last seven days and save their headlines, dates, summaries, and links in briefs/.
Restart with /exit and claude, then test it with @agent-researcher find this week's updates and save the notes.
Step 3: Add the writer and checker
Create a Sonnet writer that saves a five-item brief from the latest notes. Add an Opus checker with read-only tools to review links against the notes, dates, item count, word count, and banned words.
Run the researcher, writer, and checker in that order. Send any fixes back to the writer once, then rerun the checker. Open the source links before sending the brief.
Step 4: Schedule it weekly
In Claude Desktop, open Code, then Scheduled, then New task, then Set up manually. Paste the run instructions, enable Only on this computer, select ~/weekly-brief, and set the schedule to Mondays at 8:00 AM.
Click Run now. Review each permission request and allow recurring access for the tools the task needs. Keep Claude Desktop open and your Mac awake at the scheduled time.
P.S. Become an AI Operator here.
xAI gave Grok Bot on Mac access to Messages, Mail, Notes, Reminders, Calendar, Location, and Contacts. That lets the agent draw on messages, notes, schedules, and contact details while working on a task.
Meta launched outbound calling in Muse for paid US subscribers. The agent can place calls for tasks such as booking an appointment or asking a restaurant for a table.
Perplexity updated the browser agent in Computer and Teammates with a live view. You can take over mid-task, sign in when a site requires it, and hand control back to the agent.
🔌 Firecrawl: Launched MCP v3 with more than 3,000 tools, automatic provider switching, and parallel subtasks. Firecrawl reports 60 times higher rate limits and four times the usage allowance.
🗒️ Granola: Released MCP 2.0 with tools for writing agendas and notes, plus Agent Brief for sharing context with an AI assistant. Granola says Smart Search uses 85% fewer tokens.
🎨 Krea Bots: Opened a beta for agents that specialize in different creative tasks, remember what worked, and collaborate with teams in chat. The beta includes $1,000 in credits.
🎙️ LiquidCodec: Liquid AI’s 74-million-parameter streaming audio codec powers LFM3-Audio. It encodes audio at about 361 times real time on a single Apple M4 Max CPU thread.
THAT’S IT FOR TODAY
Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!
See you tomorrow :)
- Dr. Alvaro Cintas
✓ Full archive of premium guides with ready-to-use prompts
✓ Structured AI courses (step-by-step, start-to-finish)
✓ Every upcoming premium tutorial
✓ $500+ worth of AI tools