ChatGPT Work: Guide to OpenAI's New Agent Layer (2026)

OpenAI launched ChatGPT Work on July 9, 2026 — and it fundamentally changes what "using ChatGPT" means. Instead of answering a question and stopping, ChatGPT Work takes a goal, plans the steps, executes them in the background for minutes or hours, and hands you a finished deliverable: a spreadsheet, a slide deck, a report, or a live web app.

This guide covers every feature you need to know, how to get started with your first autonomous workflow, and the real-world results early adopters have already measured.

What Is ChatGPT Work?

ChatGPT Work is an agentic layer built into ChatGPT that runs on top of GPT-5.6 — OpenAI's newest model family. Unlike standard ChatGPT, Work is designed to:

• Connect to your existing tools and files via a unified plugin directory
• Break multi-step projects into smaller sub-tasks and execute them independently
• Deliver finished artifacts rather than conversational text
• Run on Scheduled Tasks so workflows repeat without your involvement

Think of it as the difference between asking an assistant a question versus handing a project to a colleague who delivers results.

The Four-Step Work Pipeline

Every ChatGPT Work session follows the same structure:

1. Context gathering — Work reads your connected apps, files, and workflow data via plugins
2. Planning phase — it generates a step-by-step roadmap with clarifying questions
3. Independent execution — tasks run in the background, using computer-use capabilities when needed
4. Artifact delivery — a finished file lands in your workspace, with optional scheduled updates

The key word is finished. Work does not draft a document and stop — it populates the spreadsheet, fills in the formulas, and formats the slide deck.

Plan Mode: Staying in Control

The most important feature for safe deployment is Plan Mode. Before Work begins executing any task:

1. It surfaces a step-by-step plan describing exactly what it will do
2. You review and approve (or revise) the plan
3. You set check-in points at sensitive steps so Work pauses for your judgment
4. Action-level approvals gate operations that touch external systems (sending emails, writing to CRMs, publishing files)

This approval architecture keeps human judgment in the loop without requiring you to babysit every click. Early enterprise testers described it as "the most important setting to enable on day one."

To activate Plan Mode, open the Work tab, tap the settings icon, and enable Plan before executing. For any workflow touching production data, also enable Action approvals.

Scheduled Tasks: Automating Recurring Work

Scheduled Tasks is where ChatGPT Work starts to feel like infrastructure. You can configure tasks to run:

• Once at a specific time
• Recurring on a daily, weekly, or monthly schedule
• Event-triggered when a condition occurs (a file appears, a form is submitted, a metric crosses a threshold)
• Continuously monitoring for changes, notifying you only when human decisions are needed

Example: an events team configured Work to pull registration data from Eventbrite twice weekly, generate a formatted report comparing attendance vs. target, and drop it into a shared Google Drive folder — reducing 40% of a manager's time to zero.

Sites: Building Live Web Apps

Sites is one of ChatGPT Work's most surprising capabilities. You can ask Work to build an interactive web application — a dashboard, a calculator, a form-gated report — and publish it with a shareable link inside ChatGPT.

Common Sites use cases from early adopters:

• Lead qualification scorecards — fill in inputs, get a ranked score
• Competitive benchmark dashboards — auto-refreshed from connected data sources
• Launch readiness trackers — shared across teams with live status indicators

Sites run inside ChatGPT's hosted environment, so no deployment infrastructure is required.

Integrations: 1,400+ Plugins

ChatGPT Work ships with a unified plugin directory covering more than 1,400 integrations. You connect tools using the @-mention syntax directly in your Work prompt:

Build a weekly pipeline report using @Salesforce opportunities created this week,
cross-referenced with @HubSpot engagement scores, and output as a Google Sheet.

Work auto-suggests relevant plugins as you type and can chain multiple tools in a single workflow. The most popular integration categories at launch:

GPT-5.6 Model Tiers

ChatGPT Work runs on GPT-5.6, which ships in three tiers you can select per workflow:

• Sol — flagship reasoning model for complex coding, deep analysis, and knowledge-intensive work. Highest quality, highest usage cost.
• Terra — balanced everyday model at roughly half the cost of Sol. Best for most standard Work tasks.
• Luna — most cost-efficient tier for lighter, high-volume tasks (monitoring, simple summaries, data formatting).

Selecting the right model per task is the primary lever for managing your agent-usage credits.

Enterprise Controls

For teams, ChatGPT Work ships with a full governance layer:

• Compliance API — full audit trails of every action taken by Work agents
• Workspace-level admin controls — restrict which plugins, data sources, and capabilities each user or team can access
• Auto-review mechanism — flags connected tool actions for review before they execute
• Desktop network-access policies — define which external endpoints Work can reach from the desktop app
• Spend controls — admins set credit budgets and users can request more via a workflow

OpenAI states that red team testing blocked 100% of attempts to extract protected data from connected sources.

Real-World Results

Early adopters across sales, marketing, and operations have published measurable outcomes:

These results come from Plan mode enabled with structured briefs. Unstructured prompts with no review gates produced lower-quality outputs.

Getting Started: Your First Workflow

The fastest path to a working autonomous workflow in three steps:

Step 1 — Pick a familiar, low-stakes workflow. Lead triage, competitor benchmarking, or a weekly summary report are ideal. Avoid workflows touching production databases until you are comfortable with action approvals.

Step 2 — Enable Plan Mode and Action Approvals. Review the plan Work generates before approving. Edit any steps that look wrong.

Step 3 — Run once manually, then schedule. Validate the output quality against your existing process before enabling a recurring schedule. Measure time saved after two cycles.

Limitations to Know

ChatGPT Work is powerful but not infallible:

• Outputs require verification — spreadsheets can be "polished-but-wrong." Build a review step into any financial or analytical workflow.
• No company-specific standards without detailed briefs — Work does not know your internal style guides unless you provide them as context.
• Usage is metered — longer, more complex tasks consume more agent-usage credits from your plan. Monitor spend in the admin dashboard.
• Plugin availability varies — some enterprise tools require custom connector setup by your IT team.

Conclusion

ChatGPT Work marks a genuine shift in how knowledge workers interact with AI. Rather than prompting for answers, you're now delegating for outcomes. Plan mode keeps the process auditable, Scheduled Tasks make workflows persistent, and the 1,400+ plugin directory means Work can reach the tools your team already uses.

Start with one workflow you know well, measure the time savings after two automated cycles, and then expand. The teams seeing the biggest returns are the ones who treated the first few deployments as pilots — not full automation hand-offs.

ChatGPT Work is available now to Plus, Pro, Business, Enterprise, and Edu subscribers. Free plan users get limited access via the macOS and Windows desktop apps using the Terra model.