Three things happened in three days. On May 26, xAI launched Custom Skills for Grok, a feature for building reusable parameterized tasks that the assistant can invoke on a schedule or on demand. On May 27, OpenAI rolled out governance for Skills in ChatGPT Enterprise and Edu: a dedicated admin Skills page, role-based permissions for who can create, share, install, and publish skills, and an audit surface that makes the feature deployable in regulated environments. Anthropic shipped its version of Skills as part of Managed Agents in March, with broadly similar semantics. Three labs, same word, same shape.
The pattern matters because the previous attempts at this primitive did not stick. OpenAI’s Plugins did not stick. Custom GPTs did not stick at the enterprise tier, where they got eaten by procurement and never recovered. ChatGPT’s “Actions” became internal plumbing for Skills rather than the user-facing surface. The reason “Skills” is sticking, where the previous nouns did not, is that the API shape is finally aligned with the way enterprise IT thinks about access: a Skill is a named, parameterized, callable unit of agent behavior, with an owner, a permission set, and an audit trail. You can map a Skill onto an existing IAM model without inventing a new mental category. Plugins required a new mental category. Custom GPTs required a new mental category. Skills did not, and that is the entire reason they are working.
The Clank-flavored read is that “Skills” is also just a great noun. It is concrete, it is non-threatening, it is what you teach an employee. Enterprise buyers say yes to “add a Skill” much faster than they say yes to “deploy an agentic tool registry.” The lab marketing teams figured this out independently and converged on it, which is what happens when the underlying object is real. The interesting question now is who builds the Skill marketplace. OpenAI has the workspace install base, Anthropic has the API developer base, xAI has the consumer-power-user base. None of them has the cross-vendor exchange. That is the slot somebody is about to fill, and it is going to be a louder fight than the model-tier benchmark fights have been.