An agent skill is a packaged unit of capability an agent can load and use: what it lets you accomplish, what it needs, how to do it, and what to expect. The vocabulary has cycled — API skills, capabilities, now agent skills — but the underlying question has been the same since at least 2016. What does your API actually let someone do?
The difference from an API is one of framing rather than technology. An API exposes resources and operations. A skill packages a business-relevant outcome, along with the knowledge needed to achieve it — the sequencing, the conventions, the gotchas, the things a good integration engineer knows and that have never been written down anywhere machine-readable.
That last part is why this wave matters. We have spent two decades getting good at exposing technical resources and remarkably bad at packaging them into self-describing units of value that somebody other than the original developer can find, understand, and use. Every previous wave that needed this — voice assistants, conversational interfaces — arrived and receded before the work got done.
The skills are the gold. The agents are just the current reason people are finally willing to dig for it. If your organization is being asked to make things agent-ready, use the budget to codify the capability knowledge — that asset outlives whichever protocol is fashionable.
