Lost Between Terminal Windows and AI Chat Tabs

Cmd-Tab. No. Cmd-Tab. No, that's the book repo. Cmd-`. Which of these three terminals is running the game build? Why does this chat tab say "continue" and nothing else?

You finally find the right window, type your prompt — and realize two lines later that you're in the wrong project.

Part 6 of Working for the AI — the window maze that grows around parallel agents.

The short answer#

• Running agents on several projects leaves a maze of terminals and chat tabs that all look the same.
• Hunting for the right window is a small cost paid dozens of times a day — and it invites mistakes, like typing into the wrong project.
• Open windows from the task, not the other way round: task → folder → session, every time.
• Close sessions when a task is done; a window you don't need is a window you'll mistake for one you do.

How the maze grows#

Every task opens something: a terminal for the agent, another for the dev server, a chat tab to think, an editor for the diff. Few of them get closed, because you might need it. By the afternoon:

• Terminals all look the same — black, a prompt, some scrollback.
• Chat tabs are titled after their first message.
• Nothing tells you which window belongs to which task.

So you hunt. Each hunt takes seconds, but you do it dozens of times a day, and every hunt is a small context switch. Worse, it invites the classic mistake: the right prompt in the wrong window.

Open from the task, not from the window#

Flip the order. Instead of finding a window and remembering what's in it, start from the task:

1. Pick the task.
2. Open its folder.
3. Resume its session — or start a fresh one there.

The window is created by the task, so you always know what it's for. This is the same principle as noting which session belongs to which task: the task is the index, windows are disposable.

Close windows when tasks close#

When a task is done or parked, close its windows. Nothing important should live only in a terminal's scrollback or a chat — it belongs in the repo's state file. If closing a window feels risky, that's a sign something needs to be written down first.

A good end-of-day check: could you close every window right now without losing anything? If yes, your state is where it should be.

Fewer windows, same parallelism#

You can still run agents in parallel. Just keep open only what you're actively watching — and when you want to check on the others, open them from their tasks. The agents do the parallel work; you don't need to watch every one of them at once. That's how you stop being the router for your agents.

Keep reading#

• Can't Find the Claude Session That Had the Fix?
• Context Switching Between Too Many AI Projects Kills Your Day
• Are You Working for Your AI Agent? The Copy-Paste Trap