ChatGPT Work for AI Product Managers: The Guide to OpenAI's New Agentic Workspace
TL;DR
On July 9, 2026, OpenAI launched ChatGPT Work: a new agentic mode inside ChatGPT powered by GPT-5.6. Work sits alongside Chat and Codex, connects to 60-plus external tools (Slack, Gmail, Google Drive, Salesforce, GitHub, and others), and can run multi-step tasks for hours to produce a finished spreadsheet, slide deck, report, or web app. It is bundled into Plus, Pro, Business, and Enterprise plans at no extra list price. For AI product managers, the highest-value applications are deliverable production, customer signal synthesis, and scheduled automation. This guide covers what ChatGPT Work is, the connector categories PMs use most, how scheduled tasks work, enterprise admin controls, and where it sits relative to Claude Cowork.
The AI PM Minute
One tactic to make you a sharper AI PM, twice a week. 60 seconds to read. Free.
No fluff. Unsubscribe anytime.
What ChatGPT Work Is and How It Differs from ChatGPT Chat
ChatGPT Work is not a new subscription or a separate product. It is a mode inside ChatGPT, accessible from the same interface where you already use Chat. The mental model is that Chat is for conversation and analysis; Work is for getting things done with tools.
ChatGPT Chat (existing)
A conversational interface. You paste context in, get text back. No access to external tools. No persistent work between sessions. Best for analysis, drafts, and quick questions where you are providing all the context.
ChatGPT Work (new as of July 9, 2026)
An agentic mode powered by GPT-5.6. Work connects to your external tools, pulls context from them, takes actions inside them (sending messages, updating records, creating files), and can run tasks for hours to produce finished deliverables. You stay in control: you can review, redirect, or stop a Work task at any point.
ChatGPT Codex (separate)
A developer-focused coding agent for software engineering tasks. Runs in isolated sandboxes, reads codebases, generates and runs code. Separate from Work; aimed at engineering workflows rather than knowledge work.
Work is included at no additional cost in Plus ($20/month), Pro ($200/month), Business, and Enterprise plans. OpenAI's positioning is that Work does not require a new habit: you access it inside the ChatGPT interface you already use, which reduces the adoption friction relative to installing a separate desktop application.
The Six Connector Categories PMs Use Most
ChatGPT Work ships with 60-plus connectors at launch. Most PMs will use a subset of these regularly. Below is how the connector landscape maps to the six categories most relevant to PM workflows, with specific actions available in each.
Communication (Slack, Gmail, Outlook, Teams)
Read channel history and threads, search messages, post updates to channels, draft email, send messages, and create calendar events. High-value for briefing documents, stakeholder updates, and communication that spans multiple channels.
Documents and files (Google Drive, Google Docs, Google Sheets, OneDrive, SharePoint, Notion)
Read, create, and update documents. Write to spreadsheets. Create presentations (Slides or PowerPoint). Search across file stores. The most common PM use: draft a doc in Google Docs, or pull data from a sheet and write a summary doc.
Product and engineering (GitHub, Jira, Linear, Asana)
Read issues, PRs, sprint boards, and ticket details. Create and update tickets. Link to relevant issues. Pull sprint status for status reports.
CRM and sales (Salesforce, HubSpot)
Search accounts, contacts, and opportunities. Read deal stage and notes. Useful for synthesizing customer feedback from CRM notes or preparing for executive reviews where revenue data is needed alongside product data.
Knowledge bases (Confluence, Notion, Intercom)
Read and update internal wiki pages. Search across knowledge bases. Pull customer support ticket themes. Synthesize knowledge base content into briefing docs.
Web search and live data
Work has access to real-time web search through GPT-5.6. This means tasks can pull current information (competitor announcements, recent papers, pricing pages) without needing a separate connector. Useful for competitive intelligence that mixes your internal documents with external sources.
PM Deliverables Where Work Excels
The framing from OpenAI is that Work produces "finished work": spreadsheets, slide decks, reports, and small web apps. For PMs, the deliverable types where this plays out most reliably are below.
Competitive analysis documents
Instruct Work to visit competitor websites, pull pricing and feature pages, search for recent news, and compile a structured comparison into a Google Doc or Notion page. Specify the comparison framework (feature parity, pricing model, positioning language) and the format.
Product brief or one-pager
Provide Work with a Slack thread where the feature was discussed, relevant Jira tickets, and any customer quotes. Work reads these sources and drafts a brief in your template. You edit the draft rather than starting from a blank page.
Executive update deck
Point Work at your sprint metrics in a sheet, status updates in Slack or Jira, and key decisions from a Notion doc. Work assembles a slide outline (Google Slides or PowerPoint) with the relevant data filled in. You review and adjust the narrative.
Customer research synthesis
Connect Intercom, Zendesk, or your CRM. Ask Work to extract the last 30 days of feature requests, tag them by product area, count frequency, and output a prioritized list to a sheet or doc. Removes the manual tagging step that usually falls to a junior PM or analyst.
Sprint retrospective prep
Before a retro, Work reads sprint board completions, velocity data, and any flagged blockers, then drafts a 'what went well / what to improve' starting point. The team discusses the draft rather than generating it from scratch in the meeting.
Lightweight web apps and data tools
For internal tools without UI requirements, Work can generate working web apps in a browser sandbox. A PM can build a quick calculator, a simple data viewer, or a form with no engineer required. Not suitable for production software; well-suited for internal demos and prototypes.
Scheduled Tasks and Cross-Device Sync
Work tasks can run on a schedule, and they sync across devices. You configure a recurring task on the desktop, and you can review the output from your phone. The task runs in OpenAI's infrastructure, not on your machine, so it executes whether or not you have ChatGPT open.
Scheduling syntax that works well
Work accepts natural language scheduling instructions. Specific cadences are more reliable than vague ones:
- Works well:"Every Monday at 8am Pacific, pull last week's Jira sprint data and post a summary to #product-updates in Slack."
- Works well:"Every weekday morning, check my Gmail for unread messages flagged as important and summarize them in a bullet list in my Daily Notes Google Doc."
- Avoid:"Regularly check my email." Too vague. Define the cadence, the scope, and the output destination.
Cross-device sync means Work tasks started on desktop appear in the web app and mobile. Outputs written to connected tools (docs, Slack messages, spreadsheet rows) are accessible immediately from any device. Work tasks themselves are listed in a unified "Recents" view alongside Chat history, which removes the need to track which tasks you have running.
Learn to Evaluate and Ship AI Products, Not Just Use Them
The AI PM Masterclass teaches you to make the product decisions that matter: build vs. buy, model selection, eval design, and go-to-market. Taught live by a Salesforce Sr. Director PM.
Enterprise Admin Features
For teams on Business or Enterprise plans, Work ships with admin controls that matter for IT, legal, and compliance teams. These are the ones to highlight when getting organizational approval.
Workspace-scoped Admin keys
Admins can create and manage workspace-scoped Admin keys in the Global Admin Console under Credentials. These keys work with ChatGPT and Codex administration APIs for group management, spend controls, cost reporting, and analytics. Admin keys are scoped to the workspace and expire according to the policy you set.
Spend Controls and cost reporting
Enterprise admins have visibility into per-user and per-team Work usage. Spend Controls let you cap consumption at the workspace, team, or user level. This is important for predictable budgeting in organizations rolling out Work across large teams.
Up to 120 days of task history
Enterprise and EDU plans extend Work and Codex analytics history to 120 days. This is relevant for compliance teams that need audit trails of what tasks ran, what sources were accessed, and what outputs were produced.
Group and role management
Admins organize users into teams and assign different Work capabilities by group. You can enable specific connectors for some groups (for example, Salesforce access for the product team) while keeping those connectors off for others.
ChatGPT Work vs. Claude Cowork: How to Think About Both
Both products launched or expanded in the same week of July 2026 and overlap significantly in what they can do. The relevant question is not which one is better in the abstract but which one fits your organization's existing tool stack and the workflows you most need to automate.
Existing tool commitment
ChatGPT Work
Strong if your org runs on Microsoft 365. Write tools for Outlook, Teams, SharePoint, and OneDrive are deeper in Work than in most alternatives at launch.
Claude Cowork
Strong for Google Workspace users. Cowork's Google connector covers Drive, Docs, Sheets, Gmail, and Calendar with read and write access.
Model preference and existing access
ChatGPT Work
Work is powered by GPT-5.6. If your team already pays for ChatGPT Plus, Pro, Business, or Enterprise, Work is included with no additional cost or sign-up.
Claude Cowork
Cowork runs on Claude (primarily Sonnet and Opus tiers). Available on Max, Team, and Enterprise plans. Separate subscription from Claude Chat in most cases.
Delivery format
ChatGPT Work
Work's emphasis is on finished deliverables: slides, spreadsheets, reports. The framing is 'produce a document or a file.'
Claude Cowork
Cowork's emphasis is on knowledge work delegation: reading sources, synthesizing, and writing to a destination. Both can produce documents, but the framing differs.
Connector depth
ChatGPT Work
60-plus connectors at launch, with particular depth on Microsoft 365. Web search built in for live data.
Claude Cowork
Strong connector library with enterprise-managed MCP starting with Okta, enabling zero-touch provisioning. Emphasizes enterprise governance and IT control.
For most PM teams, the practical choice is whichever tool your organization already pays for. Both are capable of the core PM workflows: briefing documents, competitive research, customer signal synthesis, and scheduled automation. Evaluate on connector depth for your specific stack, not on abstract capability benchmarks.
Go From Tool User to AI Product Decision Maker
Using ChatGPT Work and Claude Cowork is one skill. Building and evaluating AI products is another. The AI PM Masterclass covers the full stack: model selection, eval design, GTM, and the live practice of shipping AI features.
Related Articles
Before you go: get the AI PM Minute
One tactic to make you a sharper AI PM, twice a week. 60 seconds to read. Free.
No fluff. Unsubscribe anytime.