ChatGPT is OpenAI’s AI assistant. Besides the familiar chat, there’s ChatGPT Work, which takes on substantial tasks and hands back results for someone to review. Plans run from Free, Go, Plus and Pro for individuals to Business, Enterprise and Edu for organizations.
Tasks that start on their own
ChatGPT can run scheduled tasks in the background, and on eligible plans a task can start from an app event instead of the clock. A new Gmail message from a particular sender, or with a particular subject, can set one off. So can a new message in chosen Slack channels. The saved prompt decides what each run does, like summarizing every supplier email as it lands.
In Slack and Microsoft Teams, staff can mention @ChatGPT in a conversation to summarize a thread or draft a response, using tools an admin has approved.
Plugins, skills and apps
ChatGPT reaches other software through plugins, installed from one directory that ChatGPT and OpenAI’s coding tool, Codex, share. A plugin can bundle skills, which are reusable instructions for a kind of work, and MCP servers, which connect to outside systems and carry out actions there. OpenAI’s admin docs use “app” and “MCP server” for the same thing.
The Gmail plugin works with a mailbox, and the Google Drive plugin works across Drive, Docs, Sheets and Slides. The Slack plugin summarizes channels or drafts replies. OpenAI’s guides also name SharePoint and Salesforce among the work tools plugins connect to.
A plugin only sees what the signed-in account can already see. Connecting Google Drive lets its plugin work with documents that account can open, and it doesn’t unlock other files or add Drive permissions.
Plugins are also where custom GPTs are heading. OpenAI is moving custom GPTs over to plugins and will retire them. A GPT’s instructions and reference files carry across as a skill, while custom actions that called outside APIs have to be rebuilt, either on an app’s read and write actions or as a custom MCP server.
Synced content and live lookups
ChatGPT handles connected data in two ways. In a normal chat or in deep research, it processes connected data transiently, without indexing it. A synced connection indexes selected content ahead of time, and each plugin’s page says whether sync is available.
Reading and acting are separate switches. For connections that support Action control, admins can allow read-only actions or an approved custom set, and App permissions decide when ChatGPT asks before using a connection. Making a plugin available doesn’t grant access to any files or records either. The connected service’s own permissions still apply.
OpenAI’s rollout advice is to start with read actions. Before enabling writes, it suggests naming an owner for the plugin and writing down how its changes would be undone.
Workspace controls on Business and Enterprise
ChatGPT Business gives a team its own workspace with admin controls, SAML single sign-on and multi-factor authentication. Enterprise and Edu add SCIM, role-based access control, domain verification, audit logs through the Compliance API, and controls for data retention and data residency.
On Business and Enterprise plans, OpenAI doesn’t train on business data by default, and its connected-service guidance extends that to information pulled in through connected apps.
Enterprise admins can also keep company accounts inside the company workspace. Once verified-domain restrictions are on, staff can’t connect a company-domain account to a supported app from a personal ChatGPT account.
Some connections are company accounts set up by an admin. @ChatGPT in Slack or Teams uses those configured connections and doesn’t inherit any individual’s file or private-channel access. Anyone who can use @ChatGPT there can reach the tools and data it’s been given, which may differ from what they could open themselves.
The OpenAI platform
Outside ChatGPT, the OpenAI API lets developers wire OpenAI’s models into the systems a business already runs. Developers create an API key in the platform dashboard and build on the Responses API, with official SDKs for Python and JavaScript among others. Function calling lets the model request data or an action from the business’s own systems, which the integration then carries out.
The API can also call remote MCP servers directly, and by default OpenAI asks for approval before any data is shared with one. For a system on a private network, Secure MCP Tunnel connects it without opening inbound firewall ports.
Data sent to the API hasn’t been used to train OpenAI’s models since March 1, 2023, unless a customer opts in. Eligible customers can set data residency per project, which keeps stored content in a chosen region, Europe among them.
ChatGPT among the other tools
ChatGPT can reach a lot of a business through plugins and the OpenAI API. A supplier email can start a task that drafts the reply and logs the order, a Slack thread can turn into a summary for the right channel, and the API can sort incoming tickets or invoices inside the systems that already handle them.
An event-triggered task is still workflow automation, and the ROI arithmetic works out whether the hours it saves justify the setup. A plugin acts on whatever the connected system holds, so an AI readiness assessment and this piece on AI-ready foundations cover what to fix first.