Claude Code for Desktop is the BEST way to build apps with AI EVER (full tutorial)
TL;DR
Alex Finn says Claude Code Desktop has surpassed the CLI — after being a “hardcore” CLI user, he argues the refreshed desktop app is now the best vibe-coding setup because it adds project-based organization, multitasking, and a much better interface for approvals and context-switching.
The big unlock is project-organized multi-session workflows — you can run multiple sessions per project, pin the important ones, and use the yellow attention dot to jump straight to places where Claude needs approval, which Alex says made his productivity “the best it’s ever been.”
Local vs. cloud matters if you don’t want agents colliding — Alex recommends local sessions for most people, but only when each agent works on separate parts of the app; cloud agents are more flexible because changes happen on Anthropic’s servers from your GitHub repo and get merged back one at a time.
His default setup is simple: accept edits mode, normal verbosity, Opus 4.6 on high — he says you usually don’t need to manually enter plan mode because Claude switches when needed, and he prefers higher thinking depth over max mode as the best speed-quality tradeoff.
Routines are the sleeper feature because they turn Claude into a nightly code-reviewer — his example is a 9:00 p.m. scheduled routine that checks the day’s commits, reviews code, and fixes bugs, with extra upside if you connect tools like Linear to have it work through issues automatically.
His recommended workflow starts with brainstorming before coding — instead of jumping straight to implementation, he opens sessions for each project, asks Claude to first read the code and react to a feature idea, then has it plan the feature only after the idea is refined.
The Breakdown
The desktop refresh that made him ditch the CLI
Alex opens with zero hesitation: the new Claude Code desktop is “the best way to vibe code ever,” even for beginners. What changed his mind is that sessions are now organized by project, so instead of one messy stream of work, he can see multiple threads inside apps like Henry and instantly tell where his attention is needed.
Why the project-based layout actually changes how you work
He shows multiple active sessions inside the same app — one on scanner functionality, one on vibe-coding features, one on email planning — and points out the yellow dot that signals when Claude needs approval. That’s the part he’s most excited about: jumping between tasks without losing the thread, because the app finally supports the reality that he’s “doing a hundred different things at once.”
Custom panes, pinned tasks, and the local-vs-cloud tradeoff
Alex likes pinning the Tasks and Plan panes so he can always see what Claude finished, what comes next, and the high-level thinking behind it. When creating new sessions, he recommends local for most people because it’s simpler, but warns that local agents should work on very separate code paths so they don’t step on each other; cloud agents are more forgiving since they work off the GitHub repo on Anthropic’s servers and get merged back one by one.
Spinning up sessions fast — and why he loves this more than older tools
The basic move is simple: new session, choose local or cloud, choose the project folder, and Claude files it into the right project automatically. He compares this to Codex, which had already pushed project organization, and shrugs that all these tools now just copy each other’s best ideas — which, in his view, is great for users.
Chat, Co-work, and Code — plus the tax story that sells Co-work
He walks through the app’s three tabs: Chat for brainstorming, Co-work for manipulating files on your machine, and Code for actual coding sessions. The sticky anecdote here is taxes: he says a job that used to take him 10 hours became a 1-hour process after dropping credit card statements into a folder and having Claude categorize everything into a spreadsheet.
The settings he actually uses: accept edits, normal verbosity, and Opus on high
A lot of the power is tucked into the lower-left controls, but Alex’s advice is basically not to overcomplicate it. He stays in accept edits mode 95% of the time, leaves verbosity on normal because Claude “dialed in” the explanation level better than what he calls the overly verbose GPT-5.4, and prefers Opus 4.6 in high mode because it gives him better results without the slowdown of max.
Small quality-of-life upgrades: voice, slash commands, and pinned sessions
He’s not personally sold on talking to his computer — he jokes that he just likes typing on his Mac Magic Keyboard — but notes the new push-to-record option is there for people who want it. He’s more positive on pinned sessions and improved slash commands, especially the hover descriptions that finally make those commands easier to understand.
Routines and the workflow he thinks most people should copy
Routines are his favorite new feature after multitasking: scheduled tasks that can run locally or remotely, even when your laptop is closed. His example is a nightly 9:00 p.m. code-review agent that checks commits, reviews code, and fixes bugs; then he closes with his practical workflow — open one session per active project, create separate sessions for distinct features, ask Claude to first read the code and react to your idea, and only then move into planning and implementation. He draws a firm line at the end: for serious production coding, Claude Code beats OpenClaw and Hermes, which he still sees as better for prototypes.