Good Morning! Here’s what I have for you in today’s newsletter:
Claude can now send, reply to, and forward emails in Gmail on your behalf
Perplexity's Computer now works over email, just cc it on any thread
GenBio AI unveils AIDO Cell, a world model that simulates a living human cell
Build a full content studio inside Obsidian with today's five-plugin guide
4 new AI tools worth trying today
AI SKILLS
Anthropic expanded Claude's Google Workspace connector so it can now send, reply to, and forward emails in Gmail on a user's behalf, closing the gap between drafting a message and actually delivering it, with the feature available on all paid Claude plans.
Claude previously stopped at creating drafts, requiring a person to open Gmail and manually hit send, but the connector can now complete the entire process end to end.
Approval is required by default before any email goes out, though users can turn off repeated approval prompts, and on Team and Enterprise plans, owners decide whether members can enable that setting.
The update also extends to Google Drive, where Claude can now save generated files directly, working with text content though not images embedded in documents.
This closes the single biggest gap in Claude's Gmail integration, the difference between an assistant that helps you write something and one that actually finishes the task. For anyone handling routine replies, follow-ups, or forwards, this turns Claude from a drafting tool into something that can genuinely close out inbox work on its own.
AI TOOLS
ComfyUI launched Comfy MCP locally as open-source, the top-requested feature since Comfy Cloud MCP shipped in June, letting an AI agent see a user's actual local install, every custom node and every model already on disk, then fetch files and get a workflow running end to end.
The local version gives an agent direct visibility into a person's existing ComfyUI setup, meaning it works with whatever custom nodes and models are already installed rather than requiring a fresh, generic environment.
Cloud MCP keeps every capability it already had, and a job can now be pointed at either local or cloud, with the option to let the agent decide which one makes sense.
The update specifically calls out local MiniMax H3 workflows as one of the easiest use cases to get running with the new local setup.
Cloud-only tooling has always created a gap for anyone running heavily customized local setups, since a cloud agent has no visibility into what's actually installed on a specific machine. Giving an agent direct access to a real local install, custom nodes included, means it can work with someone's actual creative pipeline instead of a stripped-down generic version of it.
AI RESEARCH
GenBio AI, co-founded by Nobel laureate David Baker, unveiled AIDO Cell, a virtual cell world model capable of simulating a human cell's natural state and its response to drugs and other interventions across its full biological hierarchy, from DNA and RNA through protein to the whole-cell level.
Unlike AlphaFold, which predicts the structure of individual proteins, AIDO Cell aims to simulate how the molecular machinery of an entire cell behaves as a connected system.
The model is built as a world model rather than a single-task predictor, allowing researchers to take actions midway through a simulation and observe how the cell's state responds, rather than only generating one-shot predictions.
GenBio AI frames this as an early step toward a full digital organism, with the virtual cell serving as the foundational layer researchers can build on for future, more complex biological simulations.
Drug development has historically relied on slow, expensive trial-and-error experimentation in physical cell cultures before anything reaches human trials. A working simulator that can model how a real cell responds to an intervention offers a faster, cheaper way to test hypotheses computationally before committing to a wet lab experiment.
HOW TO AI
Your ideas live in the Notes app. Your scripts live in Google Docs. Your screenshots live in Photos, sorted by nothing. None of that is a discipline problem. It's a tooling problem. You're running a real content operation across apps that were never built to talk to each other.
This is the fix, start to finish, the actual setup, the plugins that turn Obsidian into a working system, and the real week-by-week workflow that ties it together.
Setting up Obsidian
Go to obsidian.md, download it, install it like any other app. Free, no account required. Create a new vault, then go to Settings → Community plugins and turn them on. Before installing anything, build four folders: Ideas, Drafts, Published, and Swipe File, the structure everything below plugs into.
Turn a repeated note into one click
Templater turns a note type you create repeatedly into a fill-in-the-blanks template. Install it, set a Template folder, then build one for every video idea:
---
status: idea
date: <% tp.date.now() %>
working-title:
target-length:
audience:
---
## Hook
## Core idea
## Why nowNow, whenever a new idea hits: Command palette → Templater: Create new note from template. A new note appears in your Ideas folder, date auto-filled, the rest waiting while the thought is still fresh.
See your pipeline instead of just remembering it
Kanban turns your content pipeline into a visual board backed by plain text files, no separate database. Create a board, name it Content Pipeline, and set up five columns: Ideas → Scripting → Filming/Writing → Editing → Published. Drag a card the moment you actually start writing, and the board becomes an honest picture of what's in motion.
Check before you build
If you're working with reference material constantly, this becomes one growing habit instead of five separate tricks. Every book, video, or doc gets converted on intake, into one skills folder that gets a little smarter every time you feed it something.
P.S. You can access all the AI trainings, prompts and workflows if you upgrade.
Anthropic rolled out Claude Cowork fully to mobile and web for all paid plans, letting users start a task on desktop, check progress from their phone, and pick up the finished output anywhere.
Z.ai launched the GLM-5.3 API, built for coding, defensive cybersecurity, and long-horizon agentic tasks, priced the same as GLM-5.2 and available through the official API and partner model gateways.
Perplexity brought its Computer agent into email, letting anyone send, forward, or cc [email protected] on any thread to kick off a task, with every email-triggered job running as a normal session viewable on web and mobile.
🎨 OJO: AI design agent team, turns ideas into research-backed UI designs, landing pages, and product prototypes, private beta.
🔥 Firecrawl: official connector for Claude, state-of-the-art web search for AI agents at 94.7% on SimpleQA.
☁️ Atlas Cloud: Creator Central, run the same prompt across multiple AI models side by side in one browser tab.
⚙️ NVIDIA: TensorRT Model Connect, take a Hugging Face checkpoint to native C++ inference in two commands, entirely built by OpenAI Codex agents.
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