The LLM UX problem.

Recently Theo from T3 made a video about the UX problem with LLM coding tools. The tools keep getting better, but the experience of using them is fragmented. Terminals everywhere. CLIs in tabs you can't find. Workspaces that don't cooperate across windows. Desktop environments that weren't designed for any of this.

He's right. And it's a problem I've felt every day.

The fragmentation

Our operating systems were designed around working on one thing at a time. Now I'm vibing on three projects at once. The copilot CLI is running in one terminal. Claude Code is in another. A dev server in a third. None of them know about each other. Add browser tabs, file explorers, and chat windows, and you're managing a dozen surfaces just to write code.

The new AI-first editors are exciting, but many of them optimize for the AI interaction at the expense of everything else. The result is developers jumping between tools. One app for vibing, another for review, a separate terminal for CLIs, a browser for docs. The power of AI went up. The coherence of the experience went down.

One window

My solution has been simple. Pick a full-featured IDE and put everything in it.

For me that's VS Code. One window per project. The editor, the terminal, git, the AI agent, the debugger. Everything in one place. I run the copilot CLI in a terminal pane, the Chat Panel on the side, and my code in the center. Two AI interfaces and a full editor, no app switching.

I used to keep everything separate. Dedicated terminal app, dedicated git client, dedicated everything. It felt more professional. Then I tried putting it all in one window. The friction disappeared.

Depth over breadth

Try new tools. That's how you learn what's possible, and sometimes something new genuinely changes how you work. But there's a difference between exploring and constantly switching. Every migration resets your muscle memory, your shortcuts, your configurations. Those costs compound.

The features worth having tend to show up everywhere eventually. Good ideas spread across editors fast, especially when they share a common ancestor. So explore, take note of what's good, and bring those lessons back to whatever environment you've invested in.

The key is that your home base needs to be full-featured. Agent integration, a real terminal, git integration, code diffs, debugging, extensions. You need to be able to vibe with an agent and then review its changes line by line without switching apps. Run a CLI in one pane and inspect a diff in another. Commit, push, and deploy without leaving the window. The tool has to support both modes of working, AI-assisted and manual, or you'll always end up reaching for something else.

Pick a home base

The best tool is the one you know deeply. Pick one that handles the full workflow: editing, agents, terminal, git, diffs, debugging. Consolidate into it. Learn its shortcuts, its quirks, its ecosystem. Every new feature that lands sits on top of that foundation. That's compounding knowledge, not starting over.

The UX problem is real. The fix is depth over breadth. Find a full-featured home base, go deep, and build from there.