The Power of a Skill 🛠️
Everyone has opened an AI, asked for something, gotten a response, and moved on to the next task. The problem is that's only scratching the surface. Skills turn repetitive routines into recipes the AI can run on its own, from financial closing to investigating a bug in production.
You’re probably sick of hearing about AI every single day. But whether you like it or not, it has become a commodity: it’s part of (or should be part of) the daily routine of anyone who works directly or indirectly with technology. Like anything revolutionary, whoever doesn’t use it falls behind.
What do I mean by that? If you need to travel 100 km, what’s faster: walking or driving? Driving, obviously, but only if you know how to drive. That skill is what determines whether the car helps you or gets you into trouble. Without it, you’re going to end up in a ditch.
AI extends your reach. It amplifies your knowledge and your execution speed. But the problem I see is that most people open their favorite AI, ask something or have a conversation, get a result, and move on to the next task.
That works fine for one-off cases. But there are repetitive routines that can be solved in a much faster and more structured way: the skill. Anyone who already lives in the tech world knows this concept well. This piece is for those who don’t: what exactly is this thing called a skill, and how can it help?
🍰 The Cake Recipe
Imagine you work in a company’s finance department. Every day you open a spreadsheet, take each value one by one, enter it into a system to collect the result, and paste it back into the spreadsheet. You do this for 100 rows. Then you take those rows and build a VLOOKUP against other tables to calculate some charts, arriving at yesterday’s numbers.
Notice these are repetitive steps. You do the same thing every day, and each step has its own rules: don’t copy the number if it’s negative, only consider the first two decimal places if there are more. All of this can be automated with a skill.
A skill is nothing more than a cake recipe: the AI knows exactly what to do, step by step, to accomplish a goal.
🧾 Building the Daily Closing Skill
Taking this fictional case as an example, you can go to Claude (desktop or Code, I prefer Code) and send a prompt like this:
## new skill: daily-closing
Create a skill called `daily-closing` to automate the daily financial closing process.
**What it does, in this order:**
1. Reads yesterday's spreadsheet from Google Drive, folder X (access via MCP).
2. Creates a new tab with yesterday's date, copying the structure from the "template" tab.
3. Fills in the tab following the instructions in the "filling" section.
4. Looks up, via the browser (via MCP), on site Y, the value corresponding to each transaction and fills in column H.
5. Generates an evolution chart.
**Constraints:**
- Never change or delete any tab other than the one you just created.
- Never delete files.
- Before creating the tab, check whether a tab with the same date already exists; if it does, ask before overwriting.
- If anything is unclear or ambiguous, ask before acting.From there, Claude starts building the skill, asks questions, clarifies ambiguous points, maybe asks you to confirm some detail about the spreadsheet or site Y. Only after that back and forth is the skill ready. From then on, you just call /daily-closing whenever you need to repeat the task.
🔍 Beyond the Fictional Example
This is a simple example, but it applies to a lot of things: extracting data by building dynamic queries in data warehouses, investigating issues in observability systems, and automating customer onboarding.
A skill also lets you enrich it with information and examples. That’s great when you want to give it prior knowledge of the data structure it should query, or the format of the information you expect to get back.
🩻 Bringing It Into My World
Imagine building a skill that reads a Jira ticket about a bug reported by a customer, understands the full context, and has, among its files, references to every app in the company and the role each one plays. From there, it digs through logs and traces in Grafana or Datadog, finds the error, goes to the repository of the app in question, pulls the code, and pinpoints where the bug happened. It assesses whether it was a one-off issue or whether the code needs an actual fix, replies to the internal team with the root cause and a full diagnosis, and even opens an issue with a PR suggesting the fix.
It’s wild to think this is already possible.
🧊 The Tip of the Iceberg
Even if you’re not in development, it’s worth understanding a bit more about AI and how it can help with your day-to-day work. This isn’t even the tip of the iceberg.