🧩 Anthropic Adds Mods To Claude Code

Good Morning! Here's what I have for you in today's newsletter:

Anthropic added mods to Claude Code: TypeScript add-ons that change how it works, customize the interface, or add features. You install them through plugins using /plugin in the CLI or desktop app.

Missing a context meter or want a clearer warning before a destructive command? Mods let you add those features yourself, or ask Claude to build them. Stick to mods from sources you trust, since they have the same access to your machine as Claude Code itself.

Microsoft released MAI-Transcribe-2-Streaming, its first streaming transcription model, along with the speech models MAI-Voice-2.1 and MAI-Voice-2.1-Flash, all available through Microsoft Foundry.

Nobody likes an awkward silence after asking a voice agent a question. Pairing the streaming transcriber with Flash could cut the time spent processing speech and starting a reply, leaving more room for the agent to look things up. Teams already on Azure can try the models without switching providers.

xAI added proactive suggestions to Grok Bot. Your primary Bot can now spot tasks it could handle and offer to do them, with the feature rolling out to Grok Bot users.

Figuring out what to delegate can take work in itself. Grok Bot now gives you specific suggestions to consider, so you don't have to come up with every task yourself. Receiving those suggestions won't use up your allowance, making it easier to explore what your Bot can do.

Set up your agents, connect your tools, and put recurring work on a schedule.

OpenAI Dots can keep working after you close your laptop. Its cloud computer and browser let it research, prepare documents, and coordinate tasks without using your own machine.

Start with a clear responsibility, the sources it needs, and an output you can review. Test it before putting the work on a schedule.

Pick one task you already spend time on each week. Run it through the Five-Test Rule: a specific goal, dedicated sources, a defined working style, an approval boundary, and a cadence or trigger. Treat this as a setup checklist.

The Reversibility Law: let your Dot read, research, analyze, and draft within the scope you've agreed on. Require approval before it sends messages, changes shared records, spends money, or publishes work.

Add custom permission rules under Settings → Personalization → Permissions → Custom rules for actions you always want to review first.

Messaging access and app access are separate. Connecting Slack doesn't automatically give your Dot access to your files. Have it check the sources it can open before starting the work.

To sign in to a website, open your Dot's profile, go to Computers, and open its cloud computer. Select Take over, sign in through the browser, then select Return control.

Cloud work can continue with your laptop closed. Tasks using your own connected computer need that computer online with ChatGPT open.

Create a project brief in ChatGPT Space or a document your Dot can access. Include the goal, deadline, approved claims, owners, and open decisions. Link to the source documents so it can check the details.

Review the brief, then tell your Dot to read it before working and flag anything that conflicts with it.

A single Dot can delegate work to background agents and bring their findings together. Start with a simple split: a Researcher checks the sources, a Writer drafts from approved findings, and a Reviewer checks for gaps.

Open Activity to inspect delegated tasks and correct anything that went off track.

Check the first result before making the task recurring. Specify the time, timezone, destination, and end date, then check the saved routine under Scheduled.

If you want a source monitored, ask for that explicitly. Connecting it alone won't start the checks.

P.S. You can access all the AI courses and trainings (including the full version of this one), prompts and workflows here.

Lovable added live voice conversations to Lovable apps, powered by OpenAI's GPT-Live. Builders can add talking characters or voice assistants to their apps; the demo lets users speak with Rocky from Project Hail Mary.

Comfy opened Comfy Agent to everyone in Comfy Cloud. It plans, builds, and fixes ComfyUI workflows directly on the canvas, with access to the same assets you see and support for custom or public skills. A desktop version is due in weeks.

Firecrawl launched the People Enrichment Pack in Alexandria, adding Apollo, FullEnrich, and Data Legion. Agents can use it to find buyers and work emails, search for candidates by skills and job history, and look up company size and funding.

🔬 Runway: previewed Project Continuum, a research project showing four ways to interact with a computer through real-time video interfaces.

🎥 Tavus: released Griffin, a real-time video model that 48 percent of people who spoke with it live mistook for a human.

🖼️ Black Forest Labs: released FLUX 3 Image, with pixel-precise edits across multiple rounds, bounding-box layouts, 4K output, and up to 10 reference images.

🎓 Synthesia: launched Sessions, interactive avatars that lead practice roleplays and ask survey questions, with more session types on the way.

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