OpenAI Rolls Out Custom GPTs, Skills, Workspace Agents

OpenAI Expands ChatGPT Into a Full Workflow Platform

OpenAI Academy has published a series of articles outlining a broader push to turn ChatGPT into a customizable workflow platform for businesses. Between April 10 and April 22, 2026, OpenAI detailed four connected capabilities: custom GPTs, reusable skills, research tools (search and deep research), and a new research-preview feature called workspace agents.

Custom GPTs: Purpose-Built Assistants Without Code

OpenAI announced custom GPTs functionality that lets users build purpose-built ChatGPT assistants powered by tailored instructions defining how the GPT behaves. Users can add knowledge through file uploads and enable tools such as web search, data analysis, or connected actions.

OpenAI built five example custom GPTs to illustrate the concept: ChatGPT Use Cases for Work, Professional Writing Coach, Data Analyst, Coding Assistant, and Visual Designer. The builder interface includes two tabs — Create and Configure — and OpenAI recommends setting up evaluation by writing 10 to 15 test questions with correct answers to assess a custom GPT's performance before deployment.

Skills: Reusable Workflows as an Open Standard

OpenAI Academy also released guidance on skills — reusable, shareable workflows that tell ChatGPT how to perform a specific recurring task. Skills are built using SKILL.md, a Markdown-based plain text format designed as an open standard for defining workflows.

Skills can be created directly in ChatGPT by prompting with "Build me a skill…" or uploaded from outside the platform. Once installed, ChatGPT can invoke a relevant skill automatically, or users can select one explicitly by @-mentioning it. Skills are designed to work alongside GPTs and Projects as complementary tools rather than replacements. Workspace owners retain full control over who can share and install skills, and workspace agents can also use skills within their instructions.

Search and Deep Research

OpenAI released research capabilities including a search feature that pulls the latest information from the internet directly into conversations, and a deep research feature that uses reasoning to gather, summarize, and interpret extensive information from across the web. Deep research tasks may run for 5 to 30 minutes. In enterprise environments, Workspace Owners can choose to enable or disable search functionality.

Workspace Agents Enter Research Preview

The most significant announcement is workspace agents — introduced in a April 22, 2026 Academy article — which are powered by Codex and can take on tasks such as preparing reports, writing code, and responding to messages. Workspace agents can be deployed in Slack and ChatGPT, with more surfaces described as coming soon. They support both human-triggered and schedule-triggered execution.

These agents are currently available only in research preview, across ChatGPT Business, Enterprise, Edu, and Teachers plans. In ChatGPT Enterprise, access to build agents is controlled by workspace administrators, and enterprise admins can control which connected tools and actions user groups can access via role-based access control (RBAC).

OpenAI says GPTs will remain available while teams test workspace agents against their workflows, and the company plans to make it easy to convert existing GPTs into workspace agents in the future.

On governance, the Compliance API gives admins visibility into every agent's configuration, updates, and runs. OpenAI also says admins will soon be able to view every agent built across their organization in the admin console, including usage patterns and connected data sources.

As an example of impact, Ankur Bhatt from Rippling stated that a Sales Consultant built a Sales Opportunity agent end to end without an engineering team — automating a process that previously took reps 5-6 hours per week.

Sources Differ / Open Questions

The report does not specify an exact publish date for the original workspace agents announcement itself — only for the later Academy recap article. No pricing details are provided for custom GPTs, skills, or workspace agents, and adoption data is limited to the single Rippling example. Technical distinctions between skills, GPTs, and workspace agents' underlying architecture are not detailed, nor is the configuration mechanism behind scheduled agent triggers.

Why Founders Should Care

For early-stage founders, this cluster of releases likely lowers the barrier to building internal tools without dedicated engineering resources — custom GPTs and skills suggest non-technical team members could plausibly prototype task-specific assistants directly inside ChatGPT. The reusable, shareable nature of skills (via the open SKILL.md format) may offer a practical path toward standardizing workflows as a team grows, though this remains an emerging pattern rather than a proven standard.

Deep research and search integration could reduce the need for separate research tooling in early workflows, potentially saving time on competitive analysis or market research tasks. Workspace agents' ability to run on schedules or via Slack may signal a route to automating operational tasks — such as reporting or opportunity tracking — without hiring additional engineering staff, as illustrated by the Rippling example.

However, because workspace agents remain in research preview, founders should treat current capabilities as provisional; feature stability and availability could change before general release. The planned GPT-to-agent conversion path suggests that founders experimenting with custom GPTs today may transition smoothly as agent capabilities mature, reducing the risk of wasted setup effort. Finally, the presence of admin and compliance tooling (RBAC, Compliance API) hints that even small teams may benefit from establishing basic governance practices early, before agent usage scales into something harder to track.