Something Other Than an Empty Window

Copilot is out. The fork launches, and it opens to an empty window. Now I get to put something back.

I start with a look at the current VS Code agents window. It puts AI front and center, which I like. It also has automations, chats, customizations, plugins, and MCP all in one spot. It feels like it was designed by the engineers building the platform, which makes sense, because it was. I don't edit code by hand much anymore. I tell Claude what to do and I spend my time thinking about the code. So I want the editor built around thinking about the code.

What does that look like? When the app opens I want to see the project, not a file tree:

• Journey: the Git history. Commits, branches, tags, what people have been working on.
• Architecture: the structure, the stats, the parts of the app I might want to drill into.
• Testing: what's covered and what's passing.
• A fourth card I forgot while I was dictating the prompt. It turns out to be the file explorer, rethought around what you're trying to do rather than a tree of directories.

I tell Opus 5 we are brainstorming only. No code. I have learned from past mistakes. It comes back with good pushback: build the dashboard as a real editor, not a watermark. Make the cards doors you can click into. Don't use the actual Octocat. Hide the sidebar before you rip it out.

Then I switch to Fable 5.1 and ask it to review what Opus did before we go any further. This is design work and I want the stronger model checking it. Fable flags a startup wiring problem and a contradiction in our codemap docs, which I would not have caught.

The codemap from the first section is doing its job here. Lookups are fast and the context meter barely moves. Two minutes and forty seconds in, the session has cost about a dollar.

The first build compiles and runs, and the dashboard shows up on launch. It also looks pretty rough. Claude is not good at visual design without help. I am not going to fight it. I'll fix the design in the next episode.