In partnership with
Whats up whats uppp!
You're getting this because you signed up to hear what I'm learning about AI, tools, and building stuff. I send this out every week.
Let's get into it.
What I've been using
My Claude Code Setup
I've been using Claude Code a lot these past few weeks. Not just for code, but for everything. And I realized most people don't know everything it can do.
So I put together a breakdown of how I use it.
Agent SDK
This is what blew my mind recently. The Agent SDK lets you create your own specialized agents that run inside Claude Code.
Imagine having an AI team where each one does a specific thing. One that only does research, another that only writes tests, another that reviews code. Each with its own context, its own rules and its own permissions.
The best part is they can run in parallel. While one is researching, another is writing code, another is testing.
You can spin up 5 agents at once and have them all working on different parts of the same project. When one finishes, it reports back and you can spin up another.
You can also customize how each agent behaves.
Want one that's extra careful and always asks for confirmation? Done.
Want another that just goes and does the work autonomously? Also done.
You define the guardrails for each one.
If you want to learn everything about Agent SDK, Thariq from Anthropic did a 2-hour workshop that explains it all from scratch.
MCP (Model Context Protocol)
This is probably the most underrated feature of Claude Code.
MCP lets you connect Claude to external tools. In my case I have Linear, Slack, Gmail, Firecrawl, Supabase, Sanity, and Close all connected.
Which means Claude can read a Linear ticket, search info on the web, save data in Supabase, send an email, and post in Slack. All without leaving the terminal. It's like having an assistant that has access to all your tools.
Two that I want to highlight: Firecrawl and Figma. With Firecrawl, Claude can scrape any webpage. You say "get me the pricing info from this competitor's website" and it just does it. With Figma, if you're building something, Claude can read your designs directly. No more exporting screenshots or describing what you want. It just sees the Figma file.
The setup is simple. You add a few lines to your Claude config file, restart, and suddenly Claude can talk to that tool. Most MCPs take 5 minutes to set up.
There are hundreds of MCPs available. Databases, CRMs, project management, cloud providers, you name it. You can find a directory of all of them at mcp.so. If you use a tool regularly, there's probably an MCP for it.
Doom Coding (Claude Code from your phone)
This is something I started doing recently.
You can access Claude Code from your phone using SSH. Wherever you are, with internet, you can keep working. Waiting in line, on a flight, wherever.
The setup is: Tailscale (VPN) + Termius (SSH app for phone) + your computer running at home.
There's a repo on GitHub that explains everything step by step: github.com/rberg27/doom-coding
They call it "Doom Coding" like Doom Scrolling but productive. The guy who made it has photos coding on a plane, running, and even at the club.
The reality is that with Claude Code you're not typing code, you're typing short instructions. So the phone keyboard isn't a problem.
GitHub Action (@claude in issues)
This is something I want to start using more.
You can install a GitHub Action that lets you tag @claude in any issue or PR. Claude reads it, implements the solution, and creates a PR.
The crazy part is you can have 20 open issues, tag @claude in each one, and have 20 agents working in parallel in isolated environments.
CLAUDE.md and folders
You can create a CLAUDE.md file in any folder and Claude will read it when working in that folder. It's like giving Claude instructions specific to that part of your project.
At Sidetool we have a main CLAUDE.md at the root with general rules, and then specific ones in different folders. The backend folder has one with API conventions, the frontend folder has one with component patterns. Claude picks up the right context automatically.
--resume
I used to keep a bunch of tabs open in my terminal, one for each thing I was working on. Turns out you don't need to.
Claude saves all your sessions locally. So you can run claude --resume and it picks up exactly where you left off. Same context, same conversation. No need to re-explain what you were doing.
Generally, starting fresh conversations is better for token consumption. You want to keep context as low as possible.
But there are scenarios where resuming is way better. Like debugging a workflow you built with Claude, it's always easier to debug in the same conversation where you developed it.
Or when you want to see exactly how you built something. The full history is there.
I even created a custom command /name that analyzes the whole conversation and generates a name with the most important keywords. Then I rename the session (using /rename) with that name, and finding any old conversation becomes way easier.
--dangerously-skip-permissions
Yes, it's called "dangerously" for a reason. But if you know what you're doing and you're in a safe environment, this flag lets Claude run without asking for permission on every file edit or command.
It makes everything way faster because you're not hitting "yes" every 10 seconds. Just be careful, maybe don't run it on production.
Hooks (never repeat yourself)
This one is a bit more advanced but it's crazy powerful.
Hooks let you run custom scripts before or after Claude does something.
One really useful hook that I’m using: making Claude update its own CLAUDE.md file whenever you correct it.
So if you tell Claude "don't use semicolons in this project" once, it adds that rule to CLAUDE.md automatically. Next time, it already knows. You never have to repeat the same correction twice.
There's a great video explaining this:
If you want to see my Claude Code setup with all the commands, MCPs, and configurations I use, reply to this email and I'll send it.
What caught my attention this week
Claude Cowork - "Claude Code for the rest of your work"
If you're not technical and Claude Code seems too intimidating, I still recommend you try it. But Anthropic just launched something that might be easier.
Claude Cowork is basically Claude Code but without needing a terminal. For people who aren't developers but want to use AI agents.
It works inside the Claude Desktop app. You give it access to a specific folder and Claude can read and modify files there. You talk to it in the normal chat and it does the work.
Simon Willison (who knows a lot about this) tried it and discovered that under the hood it runs a real Linux virtual machine. Meaning it's actually secure, not just in theory.
Use case examples: processing receipt photos for expense reports, organizing files, analyzing conversations.
For now it's only available for Max subscribers ($100-200/month). It's also very early, you can't even create an artifact with AI inside it like you can on the normal Claude web.
Here's the interesting part: even though it's for non-technical people, it might make them more technical.
Cowork shows the tools, commands, and code it runs. At some point you'll start asking "what's that command?" or "why did you do this?".
You'll notice the context panel and start thinking about what info the AI needs to do its job. That's basically context engineering.
People are already building custom versions of it. Which goes back to this idea of a "personal operating system", we're either going to sit in ChatGPT or Claude all day to do work, or we'll want to customize our own apps.
What I find interesting is that this shows where Anthropic is heading. Claude Code exploded, but only developers use it. Cowork is the attempt to bring that to everyone.
If you do outbound, read this
Gmail just launched AI Inbox. It uses Gemini to prioritize messages, filter out clutter, and show you what actually matters.
For regular users, this is great. Gmail might finally get good.
But if you're running cold outbound or any kind of email marketing, this changes things. AI is now actively deciding what's important and what's not. Which means you need to work twice as hard to not look like spam.
Your emails need to feel less sales-y. More personal. More relevant. The AI is reading your subject lines, your copy, your patterns. If it smells like a template, it's getting buried.
This is something to watch closely. The bar for landing in someone's inbox just got higher.
Apple chose Google for Siri
We all know Siri sucks. And Apple Intelligence hasn't exactly been a home run either.
Well, Apple finally admitted it. They're going to use Google's Gemini to power the new Siri. It's a multi-year deal, reportedly ~$1B per year.
After so much delay with Apple Intelligence, it seems like Apple gave up on doing it alone and decided they need external help to compete in AI.
The interesting part: Apple already has ChatGPT integration in Siri. It's not clear what's going to happen with that. For now they say nothing changes, but eventually they'll have to choose.
Google passed Apple in market cap last week for the first time since 2019. Things are changing.
Google Universal Commerce Protocol (UCP)
If you have an ecommerce business or just buy stuff online, this one's for you.
Google launched a new protocol for "agentic commerce."
They developed the protocol together with Shopify, Etsy, Wayfair, Target, and Walmart.
What's going to happen: you'll be talking to Gemini, tell it "I need running shoes", and you'll be able to buy directly there without leaving the conversation. Checkout with Google Pay, without opening another app.
It's like the USB-C of shopping with AI. A standard everyone will use.
AI companies are trying really hard to stay first
We saw it a couple weeks ago with Gemini 3 and GPT-5.2 dropping back to back. And now we're seeing it again with health.
OpenAI launched ChatGPT Health a few days ago. 230 million people ask about health every week on ChatGPT. Anthropic responded immediately with Claude for Healthcare.
Both let you connect medical records and fitness apps (Apple Health, MyFitnessPal, Peloton, etc.) so the AI has context about your health.
Claude has something extra: "connectors" to medical databases like PubMed, ICD-10, and CMS.
I don't think this replaces going to the doctor. But for understanding lab results, preparing questions for your appointment, or tracking health patterns, it's useful.
Both say they don't use your data to train the models. We'll see.
That's it for this week.
If you try any of these, or have thoughts on what's happening, reply. I read everything.
-Ed
Did you enjoy this newsletter? Say thanks by checking out one of my businesses:
Liked this? Sign up here to get more.
Speak fuller prompts. Get better answers.
Stop losing nuance when you type prompts. Wispr Flow captures your spoken reasoning, removes filler, and formats it into a clear prompt that keeps examples, constraints, and tone intact. Drop that prompt into your AI tool and get fewer follow-up prompts and cleaner results. Works across your apps on Mac, Windows, and iPhone. Try Wispr Flow for AI to upgrade your inputs and save time.