Posted by Hi, and welcome to Try AI for Growth, a baby podcast out of Make Space for Growth. Here I share short and sometimes surprising stories of how I’ve used technology to tackle everyday challenges. At home, at work and in business. I am your host, Sara Vicente Barreto and today, I want to tell you about how I am using Skills in Claude to teach AI how I work.
Because I realised recently that I was spending quite a lot of time repeating the same process all over again (monthly or yearly checks), and because I had been hearing a lot of buzz about skills, I decided to investigate further.
So, What Is a Skill?
Let’s start with the obvious question. What exactly is a Skill? Why do they have to make up new things all the time?
The simplest way I have found to think about it is this: a Skill teaches Claude how you want a particular type of work done. It can contain instructions, resources and even scripts that Claude can use when it recognises that particular task.
- So perhaps you always analyse a particular report in the same way.
- Perhaps your presentations always need to follow certain brand guidelines.
- Perhaps your team has a very specific process for reviewing a client proposal.
Or perhaps, like me, you have a nine-step accounting reconciliation that you really don’t want to explain from scratch every time.
You can turn that knowledge into a Skill. And then Claude can use that process again when you need it. With no special prompt needed.
For me, this immediately sounded useful. Because I had basically been creating my own version of Skills already. I just didn’t call it that.
My Old System: The Playbook
I do quite a lot of accounting reconciliations. Not the sexiest topic, but hear me out. It applies to any data checks you need to work through.
Some of these are monthly. Others happen at year-end. And many involve a fairly precise sequence of checks. Over time, I had already found a way of making Claude for Excel much more useful for these repetitive processes.
- I would create a tab in the workbook called something like “Playbook”.
- And in that tab, Claude and I would write down the process. Well, more Claude than me, really.
Step one, do this.
Step two, check this.
Step three, reconcile this against that.
If there is a difference, investigate it in this way.
First read-only. Then correct if I authorise.
And so on.
Then, when I opened a new Claude conversation, I could simply say: read the Playbook and let’s go through the process.
That worked really well. It meant I wasn’t relying on my memory, and I wasn’t having to explain the whole process again. I had effectively created an instruction manual for Claude inside my Excel file.
But it wasn’t perfect. The playbook belonged to that workbook. And I had to update this all the time. So I had to remember to copy it into new spreadsheets or projects.
And for other types of work, where I had not created a playbook but I had done something I wanted to repeat, my system was even less elegant. Sometimes I would remember that Claude and I had done a particular piece of work really well before: perhaps a client report in a particular format, or a document where I wanted to apply my branding.
But I hadn’t written down the process. So what did I do?
- I went looking for the old conversation.
- Scroll.
- Search.
- Was it this chat?
- No.
- Maybe that one?
Eventually I’d find the conversation and try to reconstruct what we had done. It worked. But there had to be a better way.
Turning Steps Into a Skill
So I decided to test Skills on one of my year-end accounting processes. This was a good test because it wasn’t a vague instruction like “make this spreadsheet better”.
It was a real process. It had nine steps. There was an order to them. There were specific checks I wanted performed.
And I already knew that the process worked because Claude and I had just gone through it together. Quite a few times. So I went through it again and, once we had completed the whole thing, I used /Skillify to turn what we had done into a reusable Skill.
And this was the moment when Skills clicked for me.
Claude essentially took the process we had just worked through and turned it into the instructions it would need to repeat that process in the future.
Underneath, there is structure to it. A Skill can contain instructions, resources and code. But importantly for me, the instructions themselves are readable.
- I can go through them.
- I can see what Claude thought the process was.
- I can check whether it had captured the nine steps correctly.
- I can change something if I need to (and it will apply to any workbook I use it in!)
And then I could save the process as a Skill. My playbook had effectively become reusable.
Claude Is Encouraging Us
Something else I noticed while working in Word and Excel was that Claude is increasingly prompting me to think this way.
When I have been doing a repeatable piece of work, I can see the option to reply as usual or use /Skillify to save the process. And I think that is quite clever.
Because often we don’t realise we have created a process. We think: I just finished the report. Claude is effectively asking: yes, but are you going to need to do that again? If the answer is yes, perhaps this shouldn’t just disappear into your chat history. Perhaps it should become a Skill.
That is a very different way of thinking about your AI conversations. Instead of every chat being disposable, some of them become the raw material for how you want work done in the future. And by the way, it makes you much less dependent on prompts!
Why This Matters Beyond My Accounting
My reconciliation is quite a specific example. But the business applications are much broader.
Imagine you produce a monthly management report.
You always want the same checks.
The same calculations.
The same comparisons.
The same questions asked of the data.
You could capture that process as a Skill.
Or perhaps you regularly prepare client reports.
You want a certain structure
A certain level of detail
Perhaps your branding and a consistent way of presenting recommendations.
That could become a Skill.
Maybe your team produces presentations, and you want everyone applying the same brand guidelines.
That could also be a Skill.
Or perhaps you have a specific way of structuring meeting notes, analysing customer feedback or reviewing a project.
Again: Skill.
The question becomes:
- What knowledge about how we work currently lives inside somebody’s head?
- And could we capture some of that knowledge so it becomes reusable?
For a small business, that is useful. For a larger organisation, I think it gets even more interesting. It reduces the risk of human error by having AI remember your process. It forces discipline in mapping out processes that we always promised we would map but never get to do. It reduces risk when someone leaves with important practical knowledge of “how we do things”.
From My Skill to Our Skill
Because a Skill doesn’t necessarily have to remain mine. On Claude’s Team and Enterprise plans, Skills can be shared with colleagues or provisioned across an organisation.
So imagine you have five people who all need to produce the same end-of-month report. Rather than each person developing their own way of prompting Claude, you can give them the process.
Or perhaps everyone in the organisation should apply the same branding rules. Same thing.
Or there is an approved process for reviewing a particular piece of data. You can capture it.
Suddenly this isn’t just about personal productivity. It becomes a way of capturing organisational knowledge, and I think that’s potentially much more powerful because every organisation has these processes.
“The way we do the monthly report.”
“The way we prepare for this meeting.”
“The way we assess this data.”
“The way we structure this document.”
Often that knowledge is sitting in a Word document nobody reads. Or, even worse, in someone’s head. And, increasingly, buried in a very good AI conversation someone had six months ago.
A Skill gives you another way to package that knowledge so Claude can actually use it when the work is happening
But Isn’t That an Agent?
Now, you might be thinking: Sara, isn’t this just an agent? I had the same question because we are now surrounded by slightly confusing AI terminology.
The distinction I find useful is relatively simple.
- A Skill is the know-how.
- An agent is the worker using the know-how.
A Skill says: when you perform this particular task, here is the process, the knowledge and the instructions I want you to follow.
An agent can have a broader goal. It may plan, make decisions about what to do next, use different tools and carry out a sequence of actions to achieve that goal.
So, if I create a Skill for my nine-step reconciliation, I am teaching Claude my reconciliation process. If I had an agent responsible for helping close my accounts at the end of every month, that agent might use my reconciliation Skill as one part of a much broader workflow.
The Skill is one capability.
The agent can decide when and how to use capabilities to get a larger job done.
You don’t necessarily need to get lost in the terminology. For me, the practical question is simply this:
Do I have a process I keep explaining to AI? If I do, perhaps it should become a Skill.
What You Can Try
So here’s the experiment I would suggest. Don’t sit down and think, “What Skill should I create?”
Instead, pay attention while you work. The next time you have a really good session with Claude, stop at the end and ask yourself:
- Will I do this again?
- Did I just teach Claude something about the way I work?
- Did we develop a sequence of steps that I would want followed next time?
- Did I correct Claude several times until it finally understood exactly what I wanted?
If so, don’t throw that work away. Try turning it into a Skill.
Start with something contained.
A monthly check. A recurring report. A document format. A branding process. A review checklist.
Something where you already know what good looks like. And importantly, review the Skill.
Just because Claude has captured a process doesn’t mean you stop thinking. Read the instructions. Check the steps. Make sure it has captured the judgement points that actually matter.
Then test it on the next piece of work. Because the point isn’t simply to save a prompt. The point is to capture a process that works. And if something did not work quite right, ask Claude to rewrite it with your changes. It is super-easy to update.
Lessons Learned
I think we can gather a few lessons today:
- If you’re repeating the instructions, capture the process. My Playbook system was already solving part of this problem. Skills simply gave me a much more reusable way to do it. If you repeatedly explain the same thing to AI, that is probably a signal.
- Your best Skills may come from work you’ve already done. I didn’t sit down and invent a nine-step reconciliation Skill. Claude and I did the reconciliation first. We refined the process. Then I captured it. For me, that feels much more useful than trying to design the perfect Skill from scratch.
- This is about consistency as much as speed. Saving time is useful, obviously. But if I have a process that I know works, I also want Claude to follow it consistently. And in an organisation, where several people may be doing the same work, that becomes even more important.
- Think about the knowledge hidden in your business. This is probably the lesson I find most interesting. We spend a lot of time talking about what AI knows. Skills made me think about the opposite question: what do I know — or what does my organisation know — that I need AI to learn?
If this makes you think of a process you keep repeating, have a look at your last few Claude conversations. There may already be a Skill hiding in one of them. And if you try creating one, I’d love to hear what process you chose. Subscribe, share the episode or leave a review if these experiments are helping you find practical ways to use AI at work.
Wrap-up
I started with Playbooks because I wanted Claude to remember how I did my reconciliations. Then I started going back through old conversations because I wanted it to remember how I created reports.
Now I’m starting to think differently. When Claude and I figure out a good way of doing something, I don’t necessarily want that knowledge to stay in the conversation where we discovered it. I want to capture it. Reuse it. And, where it makes sense, share it.
ecause perhaps one of the most useful things we can teach AI isn’t another piece of information. It’s how we actually get things done.
Thanks for listening. Until next time — keep experimenting and keep having fun.