Chat or Work? What I Learned After Running Out of ChatGPT Credits
Hi, my name’s Jeremy and I’m a ChatGPT addict. “Super-user” seems an insult to super-users. And I’m not going near ‘ninja’. I use it for client research, positioning work, proposals, content planning, file analysis, and the occasional piece of tech support. And, if we’re being honest, it’s also helping me decide what movie to watch, helped me buy a car, and negotiated a $40 saving from my internet service provider.
That level of use should have made OpenAI’s latest product changes easy to absorb. It didn’t. I’ve spent a couple of weeks trying to work out whether I should be on chatgpt.com in my browser (my default), in the desktop app (a little cold for some reason), or in the Chrome extension (very cold and I miss my familiar, friendly ChatGPT).
I’m getting to understand what each feature does. But ChatGPT and I have both been confused about where I should start a task, and whether I can simply pick it up somewhere else.
Then I ran out of Work usage, bought more credits, and ran out again. That was when I stopped treating the modes as interchangeable.
The choices became easier once I separated the kinds of context each task needs. An ongoing client project, a folder on my computer, an authenticated Chrome tab, and a public research assignment each point to a different route. I also had to nudge ChatGPT to remember that the Chrome extension is an option.
Quick Take
Start in Chat for most work, including connected Gmail, Calendar, and Drive, web research, document review, and client analysis.
Keep recurring client, business development, and internal work in Projects so the chats, files, and instructions share the same context.
Move to Work when you want ChatGPT to take a longer, multi-step assignment and execute it more independently. Work and Codex draw from the same usage allowance, and purchased credits can extend supported usage after included limits are reached.
Use the desktop app when the task needs local files, desktop apps, local folders, or a separate built-in browser session. Use cloud browser for public research that does not require your login.
Use the Chrome extension when the task depends on the Chrome session you already have open, including existing tabs, signed-in websites, or several accounts logged into the same profile.
What changed, and what did not
OpenAI’s product names sit across three layers:
‘Chat’ and ‘Work’ are modes that run on web, mobile, and desktop.
‘Projects’ hold ongoing context.
The built-in browser, Chrome extension, apps, and cloud browser give ChatGPT different ways to reach information or act on it.
A recent proposal started in my business development Project, which already held my rates, proposal rules, and portfolio material. The signed-in job listing was open in Chrome. Work compared the brief with the most relevant examples, then produced the draft in my usual format.
ChatGPT can work across every part of that job, but no single surface owns the whole workflow.
The useful question for me now is: what context does this task require, and does Chat already have it?
The routing table I use now
Chat is now my default. The table below shows what I switch to when a task needs shared project context, more autonomy, local access, an authenticated browser session, or a connected service. Work appears alongside the surfaces because it changes how ChatGPT handles an assignment.
| Route | Use it when | Main boundary |
|---|---|---|
| Chat | Most everyday work: questions, connected apps, research, review, drafting, and analysis. | Normal Chat limits apply. It does not draw from the Work/Codex usage allowance. |
| Project | The work will continue across chats and needs shared files or instructions. | Organizes context. Local and browser access come from other routes. |
| Work | You want ChatGPT to take a longer assignment and execute it more independently. | Draws from the shared Work/Codex usage allowance; purchased credits can extend supported usage after included limits are reached. |
| Desktop app | The task needs local files, desktop apps, or local folders. | Local files and outputs stay on that computer unless moved or shared. |
| Built-in browser | The task needs a separate browser session managed inside ChatGPT. | Uses its own browser state, separate from Chrome. |
| Chrome extension | The task depends on open Chrome tabs or an existing signed-in session. | Chrome only. Requires broad browser permissions. |
| Connected apps | The task needs structured access to Gmail, Calendar, Drive, or another supported service. | Capabilities depend on the connected app and its permissions. |
| Cloud browser | A public web task can run remotely or continue in the background. | Public pages only; no credentials, password managers, sign-ins, or payments at launch. |
Keep ongoing work in Projects
Projects are the home for work that will continue over time. OpenAI describes them as workspaces where related chats, files, and instructions live together.
For one recurring client content program, the Project holds the audience, messaging rules, content history, and performance benchmarks. I keep content planning, reporting, and website work in separate chats, but they share the same project context. Most day-to-day work stays in Chat; I move to Work when I want a larger assignment executed end to end.
A Project organizes context. It does not by itself give ChatGPT access to your hard drive or browser. In the desktop app, the Projects view also contains local projects connected to folders on your computer. The same name covers two different things.
Start in Chat. Spend Work usage on purpose
Work is useful when I want ChatGPT to take a longer assignment and execute it more independently. It can research, use connected sources, and create finished files.
Ordinary Chat can handle much more of my normal workflow than I first realized: connected Google Docs and Gmail, web research, client-material comparisons, screenshot troubleshooting, and positioning or messaging work.
Running out twice made the usage model impossible to ignore.
On eligible Plus and Pro plans, Work and Codex draw from the same included usage allowance. Once that allowance is exhausted, purchased credits can extend supported usage. How quickly you burn through it varies with the size and complexity of the task. You can see your credit balance and recent usage on ChatGPT web under Codex Settings > Usage Dashboard. I now check it before handing Work a bigger assignment.
I had been spending Work usage on jobs ordinary Chat could handle. Moving from the website to the desktop app does not create a fresh Work allowance; it is another surface for the same Work experience.
My rule now is simple: if Chat can access the context and I want to work through the problem with it, I stay in Chat. I move to Work when the extra autonomy is worth using the shared usage allowance.
Where the work lives still matters
Cloud Work chats now sync across web, mobile, and desktop. Local chats and local files stay on the computer where they were started or created, so check where the conversation and output live before switching devices.
The desktop app matters when the task needs local access
The desktop app adds access to local files, local folders, and supported desktop apps after you approve it. Web and mobile Work cannot directly reach files sitting on your computer.
That distinction matters when the source material is a local research folder, a set of downloaded PDFs, an existing Office file, or a website running on your computer.
The desktop app also includes its own browser. It can sign in to websites, move between several tabs, use downloads, and maintain its own browser state.
Local files and outputs remain on that computer unless you explicitly move or share them. I check that before deciding where a finished asset should live.
Built-in browser or Chrome
Use the built-in browser for a separate managed session
The built-in browser runs inside the desktop app. It supports website sign-in, several tabs, downloads, autofill, password management, and extensions. It uses its own browser state rather than your Chrome profile.
I use it when I want the browsing work contained inside the task, when I am reviewing public pages, or when a fresh browser state is useful.
Use the Chrome extension for the session you already have
The Chrome extension lets ChatGPT work with your existing Chrome profile. It can use open tabs, current logins, saved browser state, and Chrome extensions.
That makes it the practical route for a task involving a signed-in Upwork listing, WordPress page, analytics account, QuickBooks screen, or another site already open under the correct account.
Email is a good example. In Chat I can search the Gmail account connected through the app. If I need ChatGPT to check several Gmail accounts already signed in within the same Chrome profile, I use the Chrome extension because those authenticated browser sessions are already there.
Because the access is broad, I check the active Chrome profile, the account shown on the page, and the site approval before letting it continue.
Connected apps and cloud browser solve different access problems
Connected apps bring structured data and actions from services such as Gmail, Calendar, and Drive into ChatGPT. For a Gmail search, calendar check, or Google Drive file, I start in Chat with the relevant app when it is available. Google Docs, Sheets, and Slides actions are grouped under the Google Drive app.
Cloud browser handles public web tasks that can continue remotely. At launch, it cannot accept credentials, use autofill or password managers, sign in to websites, or complete payments. The task stops when a site requires one of those steps.
I use cloud browser for public research and checks that do not depend on one of my accounts. I use the built-in browser or Chrome extension when the task crosses a login screen.
Where Codex fits
For most marketing work, I stay in Chat or Work and leave Codex alone.
Where the route can still break
Publishing this post exposed a practical limit. Squarespace required an authenticated browser session. When Chrome failed to attach to the active Work thread, the handoff stopped.
The right route still needs the right access in that conversation. The same problem appears when teams re-upload brand files that belong in a Project, start cloud research that ends at a login screen, or leave a finished file on one computer.
Before I start, I now decide where the context lives, what requires authentication, whether local access matters, and where the finished work should be stored.
Start with one workflow
Pick one recurring task: a proposal review, weekly content check, campaign report, or sales follow-up.
Write down where its background context lives and which sources require a login. Note any local files or desktop apps, then choose where the final output should be saved.
Assign each part of the task to the route that fits it. You will probably find one or two points where your current process forces you to rebuild context or move information by hand.
If your team keeps rebuilding the setup for every ChatGPT assignment, I can help turn the workflow into a repeatable process.
See more First Domino insights on AI and marketing workflows. If the same problem is affecting client work across your team, see how fractional CMO support works.