ChatGPT Work

The Salesforce connector is the line that caught my eye — WinBidIQ scores federal opportunities daily, and the obvious next step is pushing the ones worth pursuing straight into a pipeline tool instead of a spreadsheet. Building that matching logic myself taught me the write side is where it gets hairy: creating/updating the right record without duplicating or overwriting something a human already touched. Following on what Santino asked below about permission control — is there a review/checkpoint before ChatGPT Work commits a write to something like Salesforce, or does it just execute once it's decided on the action?

The shift from "assistant that answers" to "agent that carries a project for hours" is ambitious and powerful if it actually works. My biggest question, echoing others in the comments, is around permission control (what it can and can't touch) and whether the phone-to-desktop handoff truly syncs state or just restarts from scratch. Without a checkpoint to pause/review before irreversible steps, I'd be a bit nervous handing it real client work.

the fact that it can actually stick with a project for hours instead of dropping context mid-task is wild, feels like the first agent that respects long workflows

The phone-to-desktop handoff is the piece I'd test first: when I start a task on mobile and resume on desktop, does the in-progress state sync through my account, or does the desktop re-run it from the top? And do Scheduled Tasks fire server-side when my machine is asleep, or only while the desktop app is open? That local-vs-hosted line is what decides whether I can actually lean on the automation instead of babysitting it.

Nice one, been waiting for this. Very interested to see how token usage is.

So we're officially from "AI assistant" to "AI coworker ". 😄 Now I just need it to survive Monday morning meetings too.

How does it handle permissions when it needs to access files or apps across my workspace, and can I scope what it can touch so it doesnt go off and do something I didnt ask for?

The hours-long autonomy is exciting, but I'd love a way to step in and review or redirect the agent mid-task without killing the whole run. Something like a checkpoint that pauses for my input before a big irreversible step would make me way more comfortable handing off real client work.

Congratulations 🎊

Really excited to hunt ChatGPT Work on Product Hunt! 🚀A few things that stood out to me:📱 Start a task on your phone and pick it up later on desktop or web.💻 Desktop can work across local files, apps, and a built-in browser.🔌 Connect tools like Slack, Google Drive, Gmail, Outlook, Teams, and Salesforce.🌐 Turn prompts into interactive dashboards, trackers, and web apps with Sites.⏰ Automate recurring work with Scheduled Tasks.👨‍💻 Codex is now built into the desktop app for coding workflows.Feels like a big step from "AI that answers questions" to "AI that actually helps move work forward."Frankly not able to keep up with all new recent model updates and releases this week 👏