What changed with ChatGPT agent
A move from ChatGPT starts with project context: files, connected accounts and check-in rules. Work’s help page separates browser and desktop permissions, so record which surface runs each step before recreating a schedule elsewhere.
Work brings connected apps, files and deliverables into ChatGPT, including logged-in browser and desktop work with permission, so someone who already uses ChatGPT for research and drafting can hand off a longer task without moving those materials into a separate product. Start there if it fits.
Model choice: OpenAI’s own models, then and now
Work offers eligible OpenAI models through a picker, with GPT-6.1 Sol rolling out alongside other documented choices. No external-provider option is documented. OperatorNest lets you choose providers or bring your own subscription or key while retaining memory and task history.
Changing providers can mean re-teaching a new tool your preferences. Portable memory reduces that work.
Where each one reaches you
Work offers connected apps and triggers through Desktop and eligible web/mobile access. OperatorNest meets you in WhatsApp, Slack, iMessage, Telegram, email or the web, and the reply returns through whichever one you used to ask. That suits a job you hand off by text.
Background work, approvals and memory
Work supports schedules and event triggers as well as longer jobs, with user-defined check-ins and browser/desktop permissions; ChatGPT’s general memory controls allow review, correction and deletion, but Work project context is separate, and OpenAI’s sources leave portable export as a question to resolve before moving that context. Ask for an export first.
OperatorNest keeps recurring tasks and monitoring running while you’re away, holds sending, paying, deleting and publishing behind an approval every time, and logs a receipt for each action. You can read, correct, export or delete memory, and none of it depends on OpenAI, xAI or any other single vendor staying the course.
Where ChatGPT Work is the better pick
Stay with Work if ChatGPT already holds your context and its models suit your longer research and drafting jobs. It also has Codex-style execution for technical work, a strength if code is part of what you’re delegating.
Switching without losing your instructions
Before moving recurring work, record its instructions, connected accounts, approval settings and project context. Re-create each schedule or trigger and check the first run. Move one task at a time, compare the results and approval behavior, and retire the old schedule once the replacement works. Cross-provider Work context import remains an open question.