Your Watch Just Dinged: CLI, Desktop and Remote Control

A complete course for developers who are serious about making AI part of how they actually work.
This lesson is about the feel of the thing. I'm working in Claude Code throughout this course (OpenAI's Codex and Google's Antigravity work-ish the same way), and I want to show you the different places you can sit while you do it. We start in VS Code, which now has its own Agents window where you pick a harness. It confuses me and it confuses a lot of people. Plenty of developers still live in VS Code because they want to see the code, so they run the Claude Code CLI there, or the extension, which I've never been a fan of. It's fine, it's a little wonky.
Then we jump into the desktop app. Anthropic has bundled Claude Code into the Claude desktop app alongside regular chat and Cowork, so you can bounce between them. I point it at Moebius, my Elixir data access library, set the model to Fable with the effort slider turned up (higher effort means it goes back and checks its own work, which costs tokens, so use it when the task needs real thinking), and dictate a code review.
I want you to go through and do a code review of this code base and tell me if it's idiomatic Elixir, and more importantly, if it is going to ever crash on someone and cause some major problems at any point. What I want to do is to avoid things like race conditions, connection lockups. I want to avoid SQL injection, anything that could make this library compromised in terms of security and make me look like a fool. And then, of course, secondarily, I want to make sure that it helps me to look good to help others.
It comes back with real findings: a connection returned to the pool while still in COPY mode, dirty connections going back into the pool, a bad number in a URL killing a connection, plus a list of security foot guns. From there I just tab into the findings and tell it:
go ahead and fix one through five with regression tests
The part I love most is remote control. I start a remote control session from the app, my watch dings, and now the same session is live on my phone and iPad. I type a quick prompt from the remote side to prove it:
what's the current version
and watch it check mix.exs in the CLI and stream the answer back to the app. You can be at a coffee shop or out to lunch with a friend, your phone goes, your run is done, you tell it what to do next and put the phone back down. Welcome to the age of agentic coding. There's a lot more to the desktop app and we'll get to it as we go, but this is the general shape of working with Claude Code day to day.
A complete course for developers who are serious about making AI part of how they actually work.