AI skills turn old prompts into reusable three-minute workflows

A useful AI workflow does not always begin with a new model or a new prompt. In a video published on August 9, Robbert van Empel shows how an old LinkedIn carousel prompt became a reusable ‘AI skill’: a process that combines instructions with files, examples, visual rules and previous feedback. The result, he says, reduced a task that used to take hours to about three minutes.

The starting point is familiar. A prompt can describe a task well, but a new chat often requires the user to repeat the same context: preferred style, fonts, images, file locations, output format and lessons from earlier attempts. That repetition is manageable once, but it becomes expensive when the task is part of a weekly or daily workflow.

In the video, the skill keeps those recurring decisions together. Van Empel feeds it an existing carousel prompt, his website, example images and a set of constraints. The skill can then use that material when asked to turn a new LinkedIn post into a carousel. It is not presented as a magic button. He still checks the output and makes the final choices. When a mistake is likely to happen again, he updates the process instead of fixing only one result.

That distinction matters for people who use AI at work. A prompt is often a useful instruction for one interaction. A skill is closer to a small operating procedure: it captures how a person or team wants a recurring task to be performed. For creators, this can preserve a visual identity. For companies, it can make workflows easier to repeat and hand over, provided the files, rules and permissions are kept current.

The video also offers a practical response to the speed of AI change. Users do not need to discard everything whenever a new tool or product appears. An old prompt library can be source material for a more durable process. The sensible starting point is one task that happens often enough to justify improvement.

The three-minute result is Van Empel’s own example, not a universal productivity guarantee. The broader lesson is more useful: automation improves when feedback becomes part of the workflow. Instead of chasing every new AI feature, users can turn one proven prompt into a small, reviewable system and improve it one decision at a time.