One Hub, Not Twelve Tabs: Setting Up the New ChatGPT for Client Work

TL;DR

• The new ChatGPT has a “work” tab that plans before it acts and connects to your other tools through connectors (the thing formerly called apps/plugins).
• Free plans get the new desktop app and most base features. Sites and the ultra model tier are paid-only.
• The real win for a solo operator isn’t the new models, it’s finally having one place to run client work, email, and research instead of forty stray chats.
• I set mine up around projects and connectors first, then picked models per task second. Here’s the setup.

Last week I had a client thread in one chat, a research dump in another, and a half-drafted email in a third. I went to find the research and opened the wrong one twice.

That’s the actual problem with how most of us use ChatGPT. Not the model. The sprawl. Every task starts a new blank chat, nothing knows about anything else, and you become the integration layer holding it together in your head. OpenAI just shipped a rebuild that takes a run at fixing this, so I spent a few days moving my real work into it.

What actually changed

There’s a new tab up top. You go from “chat” to “work,” and work behaves less like a chatbox and more like something that makes a plan before it acts, then reaches out to your connected tools to do it. If you’ve used Claude’s co-working setup, it’s that idea.

Connectors are the part worth caring about. That’s the renamed plugins/apps system, and it’s how the work tab touches your email, your files, your calendar, whatever you wire in. The naming across these products is a mess right now, so if you’re confused about connectors versus apps versus plugins, you’re not behind. They’re the same drawer with a new label.

There’s also a new family of models with a selector, an ultra mode reserved for the $200 pro plan, and a “sites” feature that builds clean hosted pages from a prompt. Sites is paid-only. The base desktop app, though, is on every plan including free, which is more than the competing tools can say.

I’ll be honest about the models: I haven’t run them long enough to tell you which one wins for what. The claims are the usual claims. What I can tell you is the workspace change matters more to my week than any benchmark.

The setup a solo operator actually uses

Here’s what I’d actually do, and what I did.

Stop thinking in chats. Think in projects. Make one project per client, plus one for your own operations (email, admin, research). A project holds context, so the client project already knows who the client is, what you’ve promised, and how they talk. You stop re-explaining yourself every morning.

Then wire two or three connectors, not ten. For me that’s email and file storage. The temptation is to connect everything on day one and end up debugging integrations instead of doing work. Connect email first, because email is the binding constraint for most of us, then add one more only when you hit a wall without it.

Model selection comes last, per task, not per identity. Fast model for triage and drafting. The heavier default for research synthesis and anything client-facing. Leave ultra alone unless you’re paying for pro and actually feeling the ceiling.

The Workflow

ONE-HUB CHATGPT SETUP (solo operator)

1. PROJECTS (context containers)
   - One project per active client
   - One "Operations" project: email, research, admin
   - Drop each client's brief, tone, and open promises into their project

2. CONNECTORS (wire 2-3, not everything)
   - Email  -> triage + drafting from inside the hub
   - Files  -> pull briefs/notes without re-uploading
   - Add a third ONLY when you hit a real wall

3. WORK TAB (let it plan)
   - Use for multi-step tasks: "read this thread, draft a reply,
     flag anything that needs my decision"
   - Review the plan before it acts. Every time.

4. MODELS (choose per task, not per ego)
   - Fast  -> triage, first drafts, reformatting
   - Default heavy -> research synthesis, client-facing output
   - Ultra -> only if you're on pro and hitting the ceiling

5. DAILY LOOP
   - Open Operations project -> run email triage
   - Move flagged items into the relevant client project
   - Draft in the client project so context stays put

The point isn’t that this is finished. It isn’t. I’m a week in and still moving things around. But the shape is right: one hub, context that persists, and me making decisions instead of playing tab roulette.

Take It Further

The packaged version of this, with the exact project structure and connector setup I’m running, is what I document in the kits. Every workflow I test lands in The Dispatch first, rough edges and all, before it’s cleaned up anywhere else.