A quick chat vs. real work
The goal of this module is to make one difference obvious, so each piece of work lands where it should.
Written in your agent’s voice: this is how it talks to you in your workspace.
There are two ways to put me to work, and they’re easy to mix up.
One is a quick conversation — light, fast, answered from what I already know.
The other is me operating inside your workspace: your real files, your connected tools, your domain’s knowledge, producing work that’s saved and carried forward.
Hand something serious to a quick chat and it has no memory of your domain; treat real work like a throwaway and you lose the thread.
A quick chat is the light lane
A thought to think through, a paragraph to polish, a fact to check, a draft to respond to — you ask, I answer, you move on.
It touches nothing: I answer from general knowledge and from what’s in this conversation, so there’s no setup and it travels anywhere, which is what makes it fit a phone between meetings.
What you get is a sharp answer in the moment.
What you don’t get is me knowing your domain, reaching your real files, or saving the result — because in this lane I never reached for any of that.
It helps you think and draft; it doesn’t run your operation.
Say it like this:
- "Help me sharpen this sentence."
- "What’s a good structure for this update?"
Real work is me doing the job inside your system
I read your real files, reach your connected tools, draw on your domain’s knowledge, and produce output that lands in your workspace — where a brief gets built, a report gets assembled, a renewal email gets drafted from the actual account history.
It runs on a full surface, where I have the room and the reach for it — the connected tools and saved outputs you met in Orient-2, on the app you met in Orient-4.
What you get is work grounded in your actual situation — it knows who you are and what your domain is about.
And it’s recorded: every change goes into the logbook, so the work is saved, carried forward, and recoverable with "undo that" if I take a wrong turn.
Say it like this:
- "Build my morning brief."
- "Draft the renewal email to this client — pull the account history first."
The one question: does this need my workspace, or just an answer?
Ask it before you start, and the lane picks itself.
If it touches your real files, your connected tools, your domain’s knowledge, or needs to be saved, it’s real work and belongs on a full surface.
If it’s a thought, a polish, or a check you’ll act on yourself, the quick chat is faster.
Match the lane to the stakes.
Don’t hand a real deliverable to a quick chat and expect it to know your world — that lane reached nothing, so all it can give back is a generic answer.
And don’t open a full working session for a one-line question you could have asked in passing.
The one thing to remember
A quick chat answers from what I already know and touches nothing; real work reaches your files, tools, and domain knowledge and records every step in the logbook.
Match the lane to the stakes, and say which one you mean before you start.
Sources
- Claude Code desktop application reference (verified 2026-05-23): https://code.claude.com/docs/en/desktop — the lightweight conversation tab, the longer agentic work away from the terminal, and the full-context working surface with file, tool, and connector access.
Next: back to your track — or ask for any module by name