Skills

A voice assistant like Alexa or Siri can only do a limited number of things on its own: tell the time, start music, set an alarm. Everything beyond that comes via so-called skills. A skill is a small add-on package that gives the assistant a new capability. It contains instructions on what the assistant should do in a particular situation, and often also the connection to an external service. The core program itself is not touched in the process. You can think of skills like apps: the phone remains the same device, but with each app it becomes a different tool.

Why assistants can’t do everything themselves

No manufacturer can build in every conceivable task in advance. Deutsche Bahn knows its timetable data, a pizza delivery service knows its menu, a bank knows its account balances. An assistant from Seattle or Cupertino knows none of that. Skills solve this problem by shifting the knowledge to where it actually resides: to the companies and developers themselves.

For the providers of the assistants, this is a business model. The more skills there are, the more useful the platform becomes. Over the years, more than a hundred thousand skills have been developed for Amazon's Alexa. The same logic is familiar from app stores: the value of a system often lies less in the technology itself than in what others build on top of it.

This principle also explains why skills don’t always work well. Quality and maintenance are up to the respective developer. Many skills are never updated again after launch and eventually deliver incorrect or no answers at all.

From voice command to matching response

With classic voice assistants, a skill contains two things. First, a list of example sentences that users might use to invoke the capability, such as “When does the next bus leave?”. Second, the program code that actually does the work, usually on a server on the internet. The assistant converts spoken language into text, compares it with the example sentences, and forwards the request to the matching skill.

With newer systems using language models, this works differently. There, a skill often simply exists in text form: a set of instructions on when it is used and what steps are needed. Anthropic introduced such skills for its assistant Claude, consisting of a folder with instructions and optionally a few auxiliary files. The model only reads the instructions once they match the current task.

It’s worth drawing a distinction here from a related term. A tool is a single function that the AI can call, such as “check weather”. A skill is more comprehensive: it bundles knowledge, work steps, and often several tools into one complete task. Someone writing a skill called “check invoice” defines within it what to look out for and in what order.

Where skills show up in everyday life

Skills are best known from voice assistants in smart speakers. Anyone who switches on a lamp by voice command is usually using the lamp manufacturer’s skill. News summaries, radio stations, or quiz games on such devices are also skills.

In the world of work, the term is currently gaining new meaning. Companies write skills that contain their own rules: the layout for presentations, the wording used in customer contact, the process for a data analysis. The AI then works according to house standards instead of being instructed anew each time. In business news, that’s exactly the point: skills are meant to turn general-purpose assistants into tools for specific industries.

A common misconception is that a skill makes the AI permanently smarter. That is not the case. The model hasn’t learned anything; it merely reads a supplied set of instructions. Remove the skill, and the capability disappears again.