Building AI skills taught me that the most valuable part is not the prompt itself, but the thinking, structure, and refinement behind it.
When people hear the word “skill” in an AI context, it can sound more technical than it really is. That was definitely true for me at first. The simplest way I think about it is this: a skill is a reusable set of instructions that helps an AI handle a repeatable task more consistently. It gives the model a clearer role, clearer boundaries, and a better sense of what a useful result looks like.
That can apply to something very small or something much more involved. A good example is article publishing work. A skill can take a draft and generate a first pass at meta keywords or a meta description, which gives you something to react to instead of making you start from scratch. That is a simple but effective example of what skills can do. They reduce friction, speed up repetitive work, and help you begin from something stronger than a blank page.
My own experience has taken that same idea into a much more layered kind of work. I have been building a connected set of skills for competitive audits, covering research, synthesis, recommendations, information architecture, and presentation prep. Working on that system taught me a lot, not just about what skills can do, but about what makes them clear, useful, and worth returning to.
Why Narrower Skills Work Better
I started with a very tempting idea: maybe one really good skill could do the whole competitive audit.
That fell apart pretty quickly.
If I asked an AI to “do a competitive audit,” there was just too much packed into that request. Was it supposed to summarize the brand, compare competitors, turn findings into recommendations, suggest sitemap changes, or draft a deck? Those are not one job. They are several different jobs, each with a different goal.
That was one of the first big things I learned: skills work better when they have a narrow, specific responsibility. Once I broke the work into stages, the outputs got more reliable. They were easier to review, easier to refine, and easier to reuse.
Where Skills Really Shine
What makes skills most useful, at least in my experience, is not that they somehow change the AI. It is that they help me handle repeatable work in a more consistent way.
That showed up for me in a few ways. Skills cut down on prompt rework, improve consistency, and make handoffs easier. I was not just trying to get one good answer from AI. I was trying to create work I could come back to, build on, and carry into the next step. Once I started thinking that way, the purpose of each skill became much clearer.
What makes skills most useful is not that they change the AI. It is that they help me handle repeatable work in a more consistent way.
Breaking the Work into Steps
Things started to click once I stopped thinking of the audit as one giant prompt and started treating it like a series of smaller steps.
It began with understanding the brand itself, then moved into competitor patterns, then recommendations and structure work, and finally into presentation prep.
One of the best decisions I made was treating the slide deck as the final presentation layer, not as part of the research itself. That helped me keep the process grounded and reminded me that skill-building is not just about generating content. It is also about putting the steps in the right order.
What Worked Surprisingly Well
One big win was getting really clear about what each skill should take in and what it should produce. Once I got sharper about that, the system became much easier to manage. The AI had less room to wander, and I had a better way to tell whether a skill was actually doing its job.
A big part of improving the skills was doing the audit work myself first, then comparing the AI’s output to my own process. When something drifted from solid UX thinking, I used that gap to refine the skill and make it clearer and more useful. But that comparison also surfaced an unexpected benefit: sometimes the AI pointed to angles or approaches I had not considered right away, which made the process feel less like automation and more like collaboration.
Another win was realizing that the work needed to live somewhere more useful than a chat thread. I wanted outputs that could be saved, revisited, shared with others, and used in the next step of the process. In my case, that often meant shaping the work so it could move into places like FigJam boards, slide decks, or other working documents. That shift made the whole effort feel less like clever prompting and more like a real workflow.
It also helped to think about how the work would actually be used. If something was going to show up in FigJam, in a deck, or in a strategy discussion, it had to be clear, organized, and easy for someone else to follow.
Building the Skill Is Only Part of the Job
Getting the skill to work was only part of the challenge. I also had to think about what it would feel like for someone else to actually use it.
That became especially clear as I started designing a skill to guide someone through the full audit workflow. I realized I was having the AI skill ask the user some setup questions too early. Where the work should go was a good example. If someone was continuing an existing audit or refining a single stage, that might already be decided. Instead of making the process smoother, those early questions just added friction.
That reminded me that a skill can sound helpful and still be frustrating to use. It also has to respect the user’s momentum. If it asks unnecessary questions or pushes everyone through the same path, it becomes harder to use.
I ran into a similar issue when I tried to make skills too broad. Bigger sounded better at first, but broader skills were harder to test, harder to trust, and more likely to drift. I also had to remember that not every workflow has the same tools available every time, so having a backup plan mattered too.
What I Would Tell Someone Starting Out
If you are building your first skills, start smaller than you think you need to.
Pick one repeatable task. Make the goal clear. Be specific about the input and the output. Then pay attention to where the friction shows up. That friction is valuable. It will usually tell you whether the skill is too vague, too broad, or is trying to do work that belongs in another step.
I would also maintain a human-review mindset from the beginning. The best skills do not replace expertise. They support it. Whether the task is small and editorial or large and strategic, the value comes from giving people a better starting point.
The Lesson I Keep Coming Back To
The biggest thing I have learned is that the strength of a skill system does not come from one brilliant prompt. It comes from clear boundaries, good sequencing, and strong handoffs between steps.
That has been the real learning journey for me. Not figuring out how to make AI do everything, but figuring out how to give it the right job at the right moment, in a way that helps people do better work.
And honestly, that is what has made skill-building feel so worthwhile. It is less about prompt tricks and more about designing better workflows. Once I started seeing it that way, the whole process made a lot more sense.
Ask Sparkbox about how AI might be valuable to your organization.
Watch this space for monthly updates on what we’re exploring, what we’ve learned, and what’s next. And if you have ideas or questions, we’d love to hear from you.
We’re hosting a webinar in February to talk about practical approaches to AI. Join us!
