Every morning, sales teams across the world manually stitch together call logs, account research, and CRM data to draft follow-up emails. This week, OpenAI showed a future where an agent does that work before the first coffee is poured.

Workspace Agents Arrive as GPTs' Successor

OpenAI has released workspace agents — shared AI assistants designed for teams — as a research preview across ChatGPT Business, Enterprise, Edu, and Teachers plans. These agents run on Codex, OpenAI's code generation engine, and execute entirely in the cloud. They keep working even when the user closes their laptop.

The agents handle multi-step workflows: report generation, code writing, message drafting, and any repetitive task a team can define. Once built, an agent can be shared across the organization via ChatGPT or Slack, and teams can iteratively improve it over time.

OpenAI published three internal case studies. The sales team built an agent that ingests call notes and account research, prioritizes leads, and drafts personalized follow-up emails. The accounting team deployed an agent that completes month-end close tasks — journal entries, balance sheet reconciliations, variance analysis — in minutes. The product team runs an agent in a Slack channel that answers employee questions, links relevant documentation, and auto-creates tickets when new issues surface.

Three Things That Make This Different from GPTs

The old GPTs were single-prompt chatbots optimized for question-answer loops. Workspace agents are a fundamentally different architecture. They handle long-running, multi-step workflows. They write and execute code. They pull data from connected apps like Google Sheets and CRMs. They maintain context across sessions.

Crucially, agents can be configured to request user approval before taking irreversible actions — editing a spreadsheet, sending an email, adding a calendar event. This guardrail makes them viable for production workflows where mistakes carry real cost.

Enterprise governance is the second major shift. Admins can now restrict which tools, data sources, and actions an agent can access at the organizational level. A dashboard tracks usage metrics: execution count, active users, and failure rates. Where GPTs were personal configurations, workspace agents ship with team-level sharing and enterprise controls baked into the design.

What Developers and Teams Feel Immediately

A team builds an agent once, then deploys it to a Slack channel for real-time requests or schedules it to run at specific times. The agent improves through conversation — each interaction refines its behavior, turning team knowledge into a reusable workflow. OpenAI provides templates for finance, sales, and marketing departments to cut initial setup time.

Training materials are available through OpenAI Academy, and detailed documentation lives in the Help Center. Existing GPTs will remain supported for now, with a migration path to convert them into workspace agents coming later.

To install a workspace agent, teams use the following command within the ChatGPT interface:

bash

Create a new workspace agent

openai agent create --name "Sales Follow-up" --template sales

For Slack integration, the setup requires a bot token and channel ID:

openai agent deploy --channel C12345 --token xoxb-xxx

Agents can be configured with approval gates using environment variables:

export APPROVAL_REQUIRED_ACTIONS="send_email,edit_spreadsheet,add_event"
export APPROVAL_CHANNEL="C67890"

The agent then runs autonomously until it hits an action requiring human sign-off, at which point it posts a request to the designated Slack channel.

The speed at which agents absorb repetitive work will determine the productivity gap between teams that adopt them and those that don't.