OpenAI evolves custom GPTs with workspace agents

What happened

On 22 April 2026, OpenAI launched workspace agents inside ChatGPT. These replace custom GPTs, the task-specific tools OpenAI introduced in 2023.

Workspace agents are available on ChatGPT Business, Enterprise, Edu, and Teachers plans. They are free until 6 May 2026. After that, credit-based billing applies.

What changed

Custom GPTs responded to individual prompts. You asked a question; it gave an answer. Workspace agents work differently. They take on jobs.

Powered by OpenAI's Codex, each agent operates in its own workspace with access to files, code, and connected tools. It pulls context from Slack, Google Drive, Linear, and other systems your team already uses. It can request approvals, follow documented workflows, and run without constant supervision.

The setup process follows a familiar pattern. Describe a recurring workflow or upload a reference file. ChatGPT defines the steps, connects the right tools, adds skills, and tests the result. Teams can then schedule agents to run on a cadence or respond to messages in Slack.

OpenAI describes this as an evolution of GPTs. That framing is modest. Custom GPTs were conversational tools. Workspace agents are autonomous workers that coordinate across systems and teams.

If the architecture sounds familiar, it should. It mirrors what Claude already does with MCP connectors and skills, packaged differently. Both platforms have converged on the same model: combine skills with connected tools, add orchestration, and let the agent run.

Two platforms, one direction

Fourteen days before this launch, Anthropic released Claude Managed Agents. That service provides composable APIs for building and deploying cloud-hosted agents. Notion, Asana, and Sentry have already built on it.

The two leading AI model providers have arrived at the same architecture within two weeks. Both now offer persistent agent sessions, connected workplace tools, scoped permissions, and autonomous multi-step task execution.

They reached this point from opposite directions. Anthropic built the composable pieces first: MCP connectors, skills, Claude Code, and Cowork. Then it launched the orchestration layer on top. OpenAI started with the chatbot and is now adding agent infrastructure underneath.

The convergence tells you something important. When both major platforms independently build the same product shape, that shape is the direction of travel. The industry has moved past the question of whether agents are the future. The answer is settled.

What this means if you run a business

If you are a business owner, operations director, or department head, here is the practical read.

AI tools are shifting from "ask a question, get an answer" to "give it a job, it handles the workflow." That changes how you plan AI adoption.

Today, your team might use ChatGPT to draft an email or summarise a meeting. With workspace agents, you describe a complete process. Qualify inbound leads against set criteria. Check the CRM. Draft a personalised response. Flag priority prospects for the sales team. The agent runs the full sequence.

This is the exact progression gecco has been building toward with every client. Our three service lines follow the same path. Start with Assistants: AI tools that help individuals work faster. Add Automations: workflows that connect your systems so data moves without manual effort. Then deploy Agents: autonomous coordinators that orchestrate assistants and automations together.

Our equation has always been Assistants plus Automation equals Agents. OpenAI has now built that same logic into its platform. So has Anthropic. The platforms are catching up to the model.

The question for your business is whether your people and processes are ready to use what the platforms now offer.

The readiness question

This is where the 80/20 reality comes in. The technology is arriving fast. Both major providers are shipping agent platforms. The tools will keep improving.

But an agent is only as good as the workflow behind it. Before you build one, you need to know which processes are worth automating. You need documentation of how work actually flows, not how the org chart says it should. You need staff who understand what agents can and cannot do.

At gecco, we see this consistently. The businesses that get real value from AI are not the ones with the best tools. They are the ones that did the groundwork. They trained their people, mapped their processes, and built confidence before adding complexity.

An agent without a prepared team behind it is an expensive experiment. An agent with a prepared team is a force multiplier.

Three things you can do this week

• If your business is on a ChatGPT Business plan, open the workspace agents feature and build one agent for a simple, repetitive workflow. Lead qualification, weekly report compilation, or feedback routing are strong candidates.
• List your top five workflows that involve moving information between tools manually. These are your automation candidates. They are also the foundation for any agent you build later.
• Ask your team what they spend their time on. The best agent ideas come from the people doing the work, not from the technology.

A note of realism

Workspace agents are in research preview. OpenAI is clear about this. The feature will evolve, pricing will change, and early agents will need iteration.

That is normal. Every major AI capability launched in the past two years has improved significantly within its first six months. The value of starting now is learning how agents work while the stakes are low and the tools are free.

Do not wait for the finished product. Start with something small and iterate.

Where gecco fits

gecco helps UK businesses adopt AI in this exact sequence. We train teams, discover the right workflows, build the assistants and automations, and bring everything together with agents. Our solutions page shows how the three service lines connect.

If you want a clear picture of where your business stands with AI adoption, take the AI readiness assessment. It takes a few minutes and gives you a tailored report with practical next steps.