Most ChatGPT users already know how to use AI for one-off tasks—like drafting, summarizing, brainstorming, or answering questions. The next phase of AI use is broader and more embedded in day-to-day work. Instead of helping with isolated moments, AI is increasingly being used to support repeatable workflows that depend on shared systems, standard handoffs, consistent outputs, and real-world constraints like timing, accuracy, and process.

That’s where workspace agents in ChatGPT fit. They’re designed to be used for repeatable workflows—work you’d otherwise do manually, re-explaining the steps each time, and copying information between tools. Learn more about workspace agents in our blog post.

If you’re new to agent building, let’s focus on the core concepts first so when you start building, you’ll know how to set up your workspace agent for consistent results.

Generally speaking, an agent is a system that carries out a task with three components: a trigger, a process that may include specialized skills, and tools or systems it can connect to.

Agents are most useful when the work is:

For open-ended thinking, brainstorming, or exploratory writing, regular chat is often a better fit—especially for one-off tasks.

Agents are also different from traditional API workflows you may have built in the past. Traditional workflows in other tools are often deterministic, meaning each step is explicitly defined, and the system follows the same path each time unless you change the logic. Agents are more probabilistic. They still operate within instructions, tools, and guardrails, but they use a model to interpret context, make bounded decisions, and adjust how they move through the work.

A helpful way to design a workspace agent is to break it into parts. Think about what you would clarify before handing work to a person: what they are responsible for, when they should begin, what should make them pause or stop, which tools and information they can use, the process they should follow, and the rules they must stay within.

An agent may require access to plugins. Learn more about

See some examples of agent breakdowns below:

The examples below can be thought of as agent workflow patterns: common, repeatable patterns of work that show up across teams and functions. Each one represents a core type of workflow an agent can carry out. The specific tools, data, and outputs may vary, but the underlying pattern stays consistent.

You may start by using agents your organization has already built. Begin by understanding what the agent is designed to do well—what tasks it supports, which tools it relies on, and the kind of output it produces.

Start with a few low-risk requests to see how it behaves. Try simple inputs first, then review the results to understand how it approaches the task.

Keep in mind that even a well-built agent still benefits from human judgment. You are often the one who knows the broader context, the stakes of the task, and what a good answer should look like.

Once you understand the basics and have identified a strong use case, you can start building your own workspace agent. You may even already have skills built in ChatGPT. Workspace agents can use skills in their instructions.

Note: in ChatGPT Enterprise, access to build agents is controlled by your workspace administrators.

Agent building works best as an iterative process. In ChatGPT, the builder is part of that loop. As you test your agent, use the conversation to understand what happened, spot issues, and refine the instructions step-by-step.

Start with a few realistic examples, including both straightforward requests and messier ones with missing context or ambiguity. This helps you see how the agent behaves and where it may need clearer instructions or guardrails.

As you test your agent, it’s expected that the first version won’t be perfect. When something feels off, there are two effective ways to improve it.

After making changes, test again to confirm the update worked.

Workspace agents are designed for shared, repeatable work. Sharing an agent gives your team a consistent way to complete a task instead of reinventing the process each time.

When you share an agent, be explicit about what it’s for. In the description, include the task it handles, when to use it, what inputs to provide, and what kind of output to expect. Shared agents work best when they’re tied to a specific, recurring workflow your team already understands. Including one or two example prompts can also give others a clear starting point and make adoption smoother.

Remember that workspace admins manage connector and feature access through role-based access control (RBAC), so teammates may need the appropriate permissions for the agent to work with systems like Slack, Gmail, or other tools.