ChatGPT Agent Mode in 2026: What It Does and Where It Stops
ChatGPT Agent mode is ChatGPT operating its own virtual computer for you — it browses websites, works with files you upload, connects to your email and document sources, fills out forms, and edits spreadsheets, then hands you the result. Operator and Deep Research are gone as separate features; both were folded into Agent mode. It’s genuinely powerful for one-off, multi-step tasks. But it’s session-based, paid-only, monthly-capped, and supervised — you kick off a task, it runs for a while, and it pauses to ask you before anything consequential. That makes it a strong project runner, not an always-on assistant.
Here’s what Agent mode actually does, the limits that matter, and where a continuous inbox agent picks up.
What ChatGPT Agent mode is
Launched in July 2025 and consolidated since, Agent mode gives ChatGPT a virtual computer of its own. Instead of just answering, it operates:
- Browses the web — navigates sites, clicks, reads, and gathers information like a person at a keyboard.
- Works with files — opens documents you upload, extracts and reorganizes their contents.
- Connects to sources — links to email and document sources to pull in context.
- Fills forms — completes and submits web forms.
- Edits spreadsheets — builds and modifies them as part of a task.
This merged in the old Operator (the browsing agent) and Deep Research (the long-form research agent) — if you’re searching for “ChatGPT Operator,” that capability now lives inside Agent mode. Tasks usually finish in 5–30 minutes, depending on complexity, and you watch the progress or step away and come back.
The limits that actually matter
Agent mode is impressive, but four constraints shape what it’s good for:
- Paid only. Available on Plus, Pro, Business, and Enterprise — not Free.
- Monthly caps. Plus gets about 40 agent messages per month; Pro about 400. Each substantial task burns from that budget, so it’s a tool you ration, not one you run constantly.
- Confirmation gates. Consequential actions require your approval. Logins and payments pause entirely and hand the browser back to you to complete by hand. You’re supervising, not delegating-and-forgetting.
- Session-based, not continuous. You start a task, it runs, it ends. There’s no listener watching your inbox or calendar to kick off the work itself. Nothing happens unless you open ChatGPT and launch a session.
None of these make Agent mode bad — they make it a supervised project runner. Perfect for “research these 20 vendors and build me a comparison sheet.” Not built for “watch my inbox and handle replies all day.”
If you want recurring prompts on a clock instead of one-off sessions, that’s a different ChatGPT feature — Scheduled Tasks — and it mostly notifies rather than acts. For continuous execution, neither one fits.
What Agent mode is great for
Used within its limits, Agent mode shines on bounded, multi-step tasks you’d otherwise do manually:
- Research projects — compiling, comparing, and summarizing across many sources.
- Data wrangling — pulling information into a spreadsheet and cleaning it up.
- Form-heavy chores — filling repetitive web forms (where login isn’t required).
- Document work — restructuring or extracting from files you upload.
The pattern: a defined job with a clear endpoint, run once, with you available to approve the risky steps. See our ChatGPT productivity guide for getting the most out of it, and what are AI agents for how this fits the broader agent landscape.
Where a continuous agent picks up
The thing Agent mode doesn’t do — run unsupervised, on its own triggers, indefinitely — is exactly where Carly lives. Carly isn’t a session you launch; it’s an assistant that works your email and calendar continuously in the cloud, across Gmail and Outlook.
The differences are structural:
- It runs on triggers, not sessions. When an email arrives, when an invite lands, on a schedule — Carly fires automatically. No task to start, no agent budget to spend, no browser tab open. Set up the workflow once and it runs.
- It acts in your inbox, end to end. Triage incoming mail, draft and send replies (with attachments), label and file, manage tasks, update your CRM, record meetings. Carly finishes the work rather than handing it back at a login wall.
- It has its own email address. Each agent gets one, so it can correspond as a colleague — something a virtual-computer session can’t be.
- No monthly run cap to ration. AI agents start at $35/month and run continuously; the Zapier-style non-AI workflow steps are free and unlimited.
- You set the rules once. Describe what you want in plain English; Carly interviews you and builds the workflow, then operates inside guardrails you define instead of pausing to ask on every step.
Agent mode is the assistant you launch for a task. Carly is the assistant that works while you’re doing something else.
ChatGPT Agent mode vs. a continuous agent
| ChatGPT Agent mode | Carly | |
|---|---|---|
| Model | Session you launch | Continuous, trigger-based |
| Runs on its own triggers | No | Yes (email, calendar, schedule) |
| Monthly cap | Plus ~40, Pro ~400 runs | No run cap (agents from $35/mo) |
| Needs supervision | Yes — pauses on consequential steps | Runs inside rules you set |
| Sends email with attachments | Not its purpose | Yes |
| Triages your live inbox | No | Yes |
| Has its own email address | No | Yes (per agent) |
| Gmail and Outlook | Connect as a source | Both, fully |
| Available tier | Paid only (no Free) | AI agents from $35/month |
| Best at | One-off multi-step projects | Ongoing inbox & ops execution |
ChatGPT Agent mode is a real leap toward AI that does things — and worth using for bounded, multi-step projects when you’re around to supervise. Just know its shape: paid, capped, session-based, and watched. If the work you want handled is the recurring inbox-and-calendar grind that should run on its own, that’s a continuous agent’s job. Start with the best AI agents for productivity, or set up a Carly agent and let it run.
Frequently Asked Questions
What is ChatGPT Agent mode?
It’s ChatGPT operating its own virtual computer to complete multi-step tasks — browsing the web, working with uploaded files, filling forms, and editing spreadsheets — then returning the result. It launched in July 2025 and absorbed the former Operator and Deep Research features. Tasks typically finish in 5–30 minutes.
Is ChatGPT Operator gone?
Yes. Operator was folded into Agent mode, along with Deep Research. The browsing-agent capability that Operator provided now lives inside Agent mode, so there’s no separate Operator product to use.
How many ChatGPT agent tasks can I run per month?
Agent mode is capped: roughly 40 agent messages per month on Plus and about 400 on Pro. It’s paid-only (Plus, Pro, Business, Enterprise) — there’s no Free access. Because of the caps, it’s best rationed for substantial tasks rather than run constantly.
Can ChatGPT Agent mode run without supervision?
Not fully. Consequential actions require your confirmation, and logins or payments pause the task and hand the browser back to you to finish by hand. It’s a supervised project runner, not an unattended agent. For unattended, trigger-based execution, see Carly.
What’s the difference between Agent mode and a continuous AI agent?
Agent mode is a session you launch for a specific task; it ends when the task does. A continuous agent like Carly runs in the cloud on automatic triggers — when an email arrives or an invite lands — and acts on its own indefinitely, with no run cap and no session to start. See what are AI agents.
More: What are AI agents · Best AI agents for productivity · ChatGPT scheduled tasks · ChatGPT personal assistant · ChatGPT alternatives
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
