ChatGPT: Workspace Agents to Boost Productivity

Introduction of Workspace Agents in ChatGPT

OpenAI has recently introduced a new feature in ChatGPT: workspace agents. These agents allow teams to create shared tools capable of managing complex tasks and long-duration workflows while adhering to the permissions and controls defined by the organization. Workspace agents represent a significant advancement of the GPT models, powered by Codex, and are designed to support various professional tasks.

The agents can handle tasks ranging from report preparation to code writing and responding to messages. Operating in the cloud, these agents continue to work even in the absence of users. They are designed to be shared within an organization, allowing for collective use in ChatGPT or Slack. This means that teams can create an agent once and improve it over time.

Using Agents for Complex Workflows

Artificial intelligence has already enabled individuals to work faster and more autonomously. However, many essential workflows within an organization rely on shared context, handoffs, and decisions between teams. Workspace agents are specifically designed for this type of task. They can retrieve context from the appropriate systems, track team processes, request validations if necessary, and advance work from one tool to another.

For example, OpenAI's sales team uses an agent to gather information from call notes and account research, qualify new leads, and draft follow-up emails directly in a salesperson's inbox. This allows sales teams to spend less time cross-referencing information and more time with clients.

Availability and Creation of Agents

Workspace agents are available in research preview for ChatGPT Business, Enterprise, Edu, and Teachers plans. To get started, simply describe a workflow that your team frequently performs, and ChatGPT will help you transform it into an agent from start to finish. Creating a workspace agent can include various tasks such as software reviews, routing product feedback, analyzing metrics, lead prospecting, and managing third-party risk.

To create an agent, just describe the task to be accomplished or attach a file. ChatGPT will then assist you in defining the steps, connecting the right tools, adding skills, and testing the agent until it works as intended. Here are some examples of agents that OpenAI has created and that your team can also create:

• Software Verifier: This agent reviews employee software requests, compares them to approved tools and policies, recommends next steps, and creates IT tickets if necessary.

• Product Feedback Router: It monitors Slack, support channels, and public forums, then transforms feedback into prioritized tickets and weekly product summaries.

• Weekly Metrics Report Generator: This agent extracts data every Friday, creates charts, writes summaries, and shares a report with the team.

• Lead Prospecting Agent: It researches incoming leads, evaluates them according to your qualification criteria, drafts personalized follow-up emails, and updates your CRM.

• Third-Party Risk Manager: This agent researches vendors, evaluates signals such as exposure to sanctions, financial health, and reputational risk, and produces a structured report.

Templates and Customization

You can also quickly get started with templates for finance, sales, marketing, and more. Each comes with built-in skills and suggested tools, allowing you to quickly set up an agent and customize it later.

Integration and Use of Agents

Workspace agents can retrieve context and act across dozens of tools. Powered by Codex in the cloud, they have access to a workspace for files, code, tools, and memory. Agents do more than respond to a prompt: they can write or execute code, use connected applications, remember what they have learned, and continue their work across multiple steps.

Workspace agents can continue to work even when you are absent. You can configure them to run on a schedule or deploy them in Slack to handle requests as they come in. For example, our product team has designed an agent that proactively answers employee questions in Slack channels. The agent responds clearly, includes links to relevant documentation, and can create a ticket when it detects a new issue. This agent helps teams unblock themselves more quickly while ensuring that no important follow-ups slip through the cracks.

Sharing and Improving Agents

Today, teams can interact with agents in ChatGPT and Slack, with other interfaces coming soon. Agents can join conversations and workflows within the tools already in use, helping teams advance work with less coordination.

Manage sharing and discover workspace agents shared by your team from the Agents tab in the ChatGPT sidebar. Knowledge is often scattered across teams and systems. Workspace agents provide teams with a way to transform this knowledge into a reusable workflow: a flow that follows the right process, uses the right tools, and can be shared throughout the organization.

For example, our accounting team created an agent that prepares key elements of the monthly close, from journal entries to balance sheet reconciliations, to variance analysis. It performs these tasks in minutes, generates working papers with the necessary source data and control totals for review, all while adhering to internal policies. The agent is available in ChatGPT for the entire team to use or integrated into Slack channels so teams can ask it questions and collaborate around its results. Because agents have memory and can be guided and corrected over conversations, they improve as teams use them. Over time, agents become a concrete means of keeping team knowledge up to date: create once, improve through use, then share or duplicate for new workflows.

Control and Security of Agents

Check the analytics of active agents in your workspace from the editor menu. When you delegate work to an agent, you maintain control. You decide which tools and data it can use, what actions it can take, and when it needs approval. For sensitive steps, such as modifying a spreadsheet, sending an email, or adding an event to the calendar, you can require that the agent request your permission before proceeding.

Once the agent is shared, analytics allow you to see how it is being used, including the number of executions performed and the number of users.

Governance and Enterprise Visibility

Workspace agents come with enterprise-level monitoring and control capabilities, so administrators can protect sensitive data while providing teams with a safe way to accelerate their work through AI. Administrators of ChatGPT Enterprise and Edu can control the connected tools and actions that user groups can access. Administrators can also manage who is authorized to use, create, and share agents. Built-in safeguards help agents stay aligned with your instructions when they encounter misleading external content, including prompt injection attacks.

The API Compliance provides administrators with visibility into the configuration, updates, and executions of each agent, allowing them to monitor and guide how agents are created and used. Administrators can also suspend agents if necessary. Soon, administrators will also be able to see all agents created within their organization in the admin console, including usage modes and connected data sources.

Early User Feedback

Early users testing workspace agents are already seeing more consistent results and saving time to focus on higher-value tasks. One user mentioned that the most challenging aspect of creating an agent is not the model itself, but rather the integrations, memory, and user experience.