OpenAI Now Lets Teams Build Custom AI Bots That Work Autonomously

OpenAI has officially launched “workspace agents” — cloud-based AI bots that teams can build, share, and deploy to autonomously handle business tasks across tools like Slack, Gmail, and the open web. The feature, announced on April 22, 2026, is available to users on OpenAI’s Business, Enterprise, Edu, and Teachers plans, marking the company’s most significant push yet into enterprise AI automation.

What Are OpenAI Workspace Agents?

Workspace agents are a new category of AI-powered bots that live inside ChatGPT and can perform multi-step business workflows without constant human oversight. Unlike standard ChatGPT conversations, these agents are designed to operate continuously in the cloud, gathering context from enterprise systems, following predefined team processes, requesting human approval at critical junctures, and keeping work moving across multiple tools.

OpenAI provided several concrete examples of what these agents can do:

• Product feedback agent: Scours the web for customer opinions about a company’s products, compiles findings into a structured report, and automatically delivers it to a designated Slack channel.
• Sales follow-up agent: Monitors Gmail for incoming leads, drafts personalized follow-up emails, and queues them for human review before sending.
• Research agent: Gathers competitive intelligence from public sources, organizes data into actionable briefings, and shares them with relevant team members.

The key differentiator is that these agents are built to be shared within organizations. As OpenAI stated in its announcement blog post, “teams can build an agent once, use it together in ChatGPT or Slack, and improve it over time.” This collaborative approach transforms AI agents from individual productivity tools into institutional infrastructure.

The Evolution From GPTs to Workspace Agents

For those familiar with OpenAI’s product history, workspace agents represent a natural evolution — and likely the successor — to the company’s “GPTs” feature, which launched at DevDay in November 2023. GPTs allowed users to create custom chatbots with specific instructions, knowledge files, and capabilities. However, they remained largely conversational tools: you talked to them, and they responded.

Workspace agents go significantly further. They can:

• Operate autonomously in the background without active user sessions
• Integrate directly with third-party business applications
• Be shared and collaboratively improved across entire organizations
• Maintain persistent state and context across multiple interactions
• Trigger actions based on events rather than just responding to prompts

OpenAI has confirmed that workspace agents are an “evolution” of GPTs and that “GPTs will remain available while teams test workspace agents with their workflows.” The company also plans to “make it easy to convert GPTs into workspace agents” in the near future, suggesting a deliberate migration path from the older format.

The Competitive Landscape: Why OpenAI Moved Now

The timing of this announcement is no accident. The AI agent space has become intensely competitive, with multiple major players racing to define what autonomous workplace AI looks like.

OpenAI is facing increased competition from Anthropic, which offers its own Claude Cowork agent that can complete tasks using files from your computer, as well as a separate platform for making autonomous agents.

Perhaps the most intriguing development is OpenAI’s connection to OpenClaw — the AI agent formerly known as Clawdbot and Moltbot that went viral for its tagline as the “AI that actually does things.” OpenClaw’s founder, Peter Steinberger, has since joined OpenAI, bringing his hands-on agent-building expertise directly into the company. While OpenAI hasn’t explicitly confirmed the connection, the timing and capabilities suggest significant cross-pollination of ideas.

Meanwhile, Anthropic has been aggressively expanding its agent ecosystem. Claude Cowork, launched as part of Anthropic’s broader enterprise push, allows users to delegate tasks that involve local file access and multi-step workflows. Anthropic has also released a dedicated platform for building autonomous agents, positioning itself as a direct competitor in the enterprise automation space.

Google, Microsoft, and a host of startups are also investing heavily in AI agents, making this one of the most contested battlegrounds in the current AI wave.

How Workspace Agents Work: A Technical Deep Dive

While OpenAI hasn’t released full technical documentation for workspace agents, several key architectural principles are clear from the announcement and related materials:

Cloud-Native Execution

Unlike GPTs, which required an active user session in ChatGPT, workspace agents run in the cloud. This means they can operate on schedules, respond to triggers, and maintain continuity even when no human is actively interacting with them.

Multi-Tool Integration

Agents are designed to work across a range of business applications. The examples provided — Slack, Gmail, web search — suggest a growing ecosystem of integrations that will likely expand over time. The architecture appears to support both reading from and writing to these external systems.

Human-in-the-Loop Design

OpenAI has emphasized that agents can “ask for approval when needed,” reflecting a design philosophy that keeps humans involved at critical decision points. This approach balances automation with accountability — a crucial consideration for enterprise adoption.

Collaborative Improvement

The ability to share agents within organizations and “improve them over time” suggests a feedback loop where agents learn from usage patterns and team input, becoming more effective the more they’re used.

What This Means for Businesses

For organizations already using ChatGPT on business or enterprise plans, workspace agents represent a significant new capability. Here are the most immediate implications:

Reduced Manual Overhead

Tasks that currently require employees to manually gather information from multiple sources, format reports, and distribute them could be automated with workspace agents. The product feedback agent example alone could save dozens of hours per week for product teams.

Standardized Processes

By encoding team workflows into agents, organizations can ensure that critical processes are followed consistently, reducing the risk of human error or oversight. This is particularly valuable for compliance-heavy industries.

Democratized Automation

Workspace agents lower the barrier to creating business automation. Instead of requiring dedicated engineering resources to build custom integrations, teams can configure agents through ChatGPT’s interface — making automation accessible to non-technical staff.

Potential Concerns and Limitations

Despite the promise, there are important considerations for organizations evaluating workspace agents:

• Data privacy: Agents that access multiple business systems raise questions about data handling, retention, and compliance with regulations like GDPR and HIPAA. OpenAI has not yet published detailed documentation on how workspace agent data is secured.
• Availability: The feature is currently limited to Business, Enterprise, Edu, and Teachers plans — excluding individual users and smaller organizations on free or personal plans.
• Integration depth: While the examples show Slack and Gmail integration, the full list of supported applications and the depth of those integrations remain unclear.
• Reliability: Autonomous agents that interact with external systems can fail in unexpected ways. Organizations will need robust monitoring and fallback procedures.

The Bigger Picture: Agents as the Next Computing Platform

Workspace agents are part of a broader industry shift toward agentic AI — systems that don’t just answer questions but take actions. This shift has profound implications for how we think about software, work, and organizational design.

If traditional software tools are instruments that humans play, AI agents are more like employees that humans manage. This changes the fundamental relationship between people and technology. Instead of learning to use a tool, managers will need to learn to configure, supervise, and evaluate AI agents — a distinctly different skill set.

OpenAI’s move to make agents shareable and collaboratively improvable within organizations suggests the company envisions a future where teams maintain libraries of agents, each optimized for specific workflows, that collectively form an “AI workforce” alongside human employees.

What’s Next for OpenAI Workspace Agents

Several developments are worth watching in the coming months:

• GPT migration tools: OpenAI has promised an easy path to convert existing GPTs into workspace agents, which could rapidly populate the ecosystem with functional agents.
• Expanded integrations: The current examples cover a limited set of tools. Expect a growing marketplace of agent integrations for CRM systems, project management platforms, databases, and more.
• Pricing changes: Autonomous agents that run continuously may require different pricing models than per-session ChatGPT usage.
• Competitive responses: Anthropic, Google, and Microsoft are likely to announce their own enterprise agent features in response.

Bottom Line

OpenAI’s workspace agents represent a meaningful step toward the vision of AI as an active participant in business workflows rather than a passive tool. For teams on eligible plans, the ability to build, share, and deploy autonomous agents directly within ChatGPT could significantly reduce the friction between identifying a repetitive task and automating it.

The feature also signals OpenAI’s recognition that the real value of AI in enterprise settings lies not in individual productivity boosts but in organizational-level automation — where agents become shared infrastructure that compounds in value as more people use and improve them.

Whether workspace agents live up to this promise will depend on execution: the depth of integrations, the reliability of autonomous behavior, and the safeguards around data and decision-making. But the direction is clear, and the race to define the future of AI-powered work is well underway.

Source: The Verge — “OpenAI now lets teams make custom bots that can do work on their own” by Jay Peters, April 22, 2026

📖 Related: Anthropic Just Launched Cowork — And It Might Be the AI Agent That Actually Works

📖 Related: Google Meet’s AI Note-Taker Is Breaking Out of the Virtual World

📖 Related: Anthropic Launches Cowork: The Claude Desktop Agent That Works in Your Files Without Any Coding