AI skills start on laptops. Enterprises inherit the mess.

It's not feasible to expect a system administrator to fully comprehend how every skill within an organization should be optimally functioning, according to Sagar Batchu, the founder and CEO of Speakeasy. This challenge escalates as enterprises integrate AI agents and developers compile extensive collections of skills, encompassing deployment runbooks, code review checklists, style guides, incident procedures, and internal workflows.

These skills often originate as personal experiments before proliferating across repositories, local machines, Slack threads, and plugins. Moreover, they frequently lack clear ownership or a dependable method to ascertain if the skill being utilized is the latest version. Speakeasy unveiled Skills Management on Friday, a system intended to treat skills as centrally registered enterprise artifacts, complete with immutable versions, scoped access, and visibility into their usage.

This comes after developers began creating sets of prompts paired with configurations for agents to employ, eventually formalizing an Agent Skill specification for a skill file in Markdown format with some YAML frontmatter. Speakeasy emphasizes that skills are the optimal means of conveying company-specific knowledge to AI agents.

However, these skill libraries are proliferating at a rate that outpaces the companies' ability to manage them. A mid-sized fintech client of Speakeasy recently identified over 500 distinct AI skills in use within their organization. This is due to the inherently personal nature of AI experimentation; interacting with a chatbot feels private, similar to the manner in which Claude Code communicates.

Consequently, sharing prompts with the team might not occur promptly, leading to a significant amount of "reinventing the wheel" as employees develop similar skills without awareness of others' efforts. To address this issue, Speakeasy initiates by treating a skill as a newly registered enterprise artifact, which they term Skills Management.

Skills Management manages skills much like git handles code, offering a registry of immutable versions and a unified perspective of what exists. Therefore, when an existing skill is edited, a new version is generated. However, Speakeasy also incorporates observability, enabling the enterprise to identify which skills are most utilized and how frequently various departments duplicate efforts.

Furthermore, if a skill underperforms—meaning the agent employing it encounters issues while executing—the LLM judge can examine it and propose enhancements. Skills Management necessitates adherence to the aforementioned Agent Skills specification when generating a skill file, allowing the frontmatter name to serve as the canonical reference.

Disregarding attempts to enter the same content under a different name is a key feature. Immutable versions and access control are established, enabling the formation of roles around who can create and curate skills versus who can consume them. This combination of monitoring skills through Speakeasy's control plane as agents utilize them proves superior to merely collecting skill files within a git repository.

Validating skills against the Agent Skills specification is a prudent approach. Markdown, being essentially English with some symbols, is a sensible choice. However, some optional entries in the frontmatter, such as "license," could benefit from an enterprise template to streamline the process. Sagar informs The New Stack that such company-specific templates have not yet been implemented.

Nevertheless, companies could develop a skill dedicated to validating skills, which could be integrated into the LLM-as-judge process. The metadata surrounding each skill enables administrators to swiftly identify which skills are maintained (last updated) and their frequency of use, as well as their distribution among team members.

Therefore, would it be feasible for a human (or process) to prevent the latest detrimental change from spreading? Sagar confirms that Speakeasy can track metadata for each skill, allowing administrators to ascertain which skills are maintained and their usage frequency, along with the spread among team members. Additionally, Speakeasy is actively developing auto-categorization features.

This aspect of the Speakeasy system is notably intelligent because it recognizes how a skill commences on a developer's laptop and must be promoted throughout the enterprise to achieve full adoption, thereby exposing more complexities regarding roles. Human intervention will likely be required to manage the various responsibilities involved.