How AI skills are quietly automating my workday

A year ago I was managing my week from a dozen different dashboards. Hubspot for tracking sales. Slack for what’s happening. Notion for what we decided. PostHog for website and product analytics. Some spreadsheets for priorities. I think this is how most workdays still look today.

Today, I open one chat on Monday morning and ask: “What should I focus on this week?”

And I get a genuinely good answer. Not a generic one. A prioritized list based on what happened in Slack last weeks, what’s in my knowledge base, what projects are moving. It takes about 3 minutes. Then I can continue in the same chat: “give me an overview of website traffic over the weekend,” “how many signups do we have in the product,” “do we have any peaks or drops that require my attention.”

This is what AI skills unlocked for me, and I want to walk through how.

So what are skills, actually?

If you’ve been following the AI wave, you’ve probably heard this word thrown around. It confused me at first too. Feels a lot like “prompts,” which have been around since ChatGPT day one.

The difference is meaningful though.

Skills are sets of instructions, commands, and context that your AI can access and use automatically when it recognizes a relevant task. You don’t have to remember to feed them in every time. AI can also update and improve them on the go. It calls them when it needs them.

Four things make skills genuinely different from regular prompts:

They activate themselves. If I ask “draft an email to our investors,” my AI checks if it has a skill for that and uses it. With plain prompts, I’d need to remember to paste in my tone guidelines every single time. Minor inconvenience in isolation, significant friction multiplied across a week. It also brings you way closer to agentic and automated workflows.

They improve over time. After I run a skill, I can tell the AI: “learn from what I just corrected and update the skill.” So the next time it runs, it’s closer to what I actually want. I use this constantly with my WordPress publishing flow as I described in my previous post. The first time took a long conversation with a lot of supervision and iterations. Now it knows exactly how I want posts formatted, which callouts and FAQs to include exactly. It just does it.

They turn into real capabilities. This is the big one. When you start combining skills with tools, especially ones that connect to your apps via CLI, you stop clicking through interfaces entirely. You describe outcomes. The skill handles execution.

They’re efficient with your LLM’s context. When you paste a long prompt into a chat, it takes up your entire context window upfront. That’s consuming your AI’s working memory which is a very critical asset right now. So skills work differently by design. The AI sees a short description of each skill at startup, and only loads the full instructions when it actually needs them. So instead of dumping everything in at once, your context stays clean for the actual work.

The skill I use every Monday morning

At the end of each week, I ask Desktop Commander to run my “knowledge base update” skill.

Here’s what happens: DC connects to my Slack, finds my channels, and scrapes relevant messages from the past week. Then it opens my local knowledge base, a folder of structured notes on my machine, learns how it’s organized, and adds new company information to the right sections.

So by Monday, when I ask for my weekly priorities, the AI isn’t guessing. It’s pulling from an updated knowledge base that includes everything that happened at the company last week, plus any longer-term strategic context already in there.

The priority list it creates is surprisingly good.

The CLI skill that made me a sort of developer

Another mindblowing one was built for me by our CTO. It runs a local version of our app, lets me make changes, and creates pull requests for him to review.

The skill handles starting the local version of the application, checks app health, looks at errors, and restarts services if something is stuck. Instead of remembering many technical commands, it follows a repeatable process for app operations.

I open a chat, describe what I want to change or build, and the skill handles the rest: running the local environment and starting the app for me. I then make code changes, review immediately how it looks in the product and if all is good, create a PR. My CTO reviews and merges. No developer in the middle.

With Opus 4.6 and GPT-5.3 the quality is getting genuinely good. I’m building real features in a real app used by hundreds of users daily. I can’t wait to see where this is in 6 months.

What this has done to my actual workday

My workday now mostly happens in one tool. I ask questions, define outcomes, check results, iterate. That’s it.

I used to jump between dashboards. Now I describe what I need and the right skill handles execution, whether that’s pulling from Slack or Notion, pushing stuff to GitHub, publishing a blog post, or updating my knowledge base.

The skills that have changed my workflow most:

• Product and traffic insight generation (PostHog + instructions on how analytics is set up in our app and website)
• General app or website development (instructions and tools that turn idea into local build into pull request)
• Weekly knowledge base update (Slack into local notes)
• WordPress publishing (draft .docx file into formatted post, published)
• Accessing call transcripts (instructions to access Fireflies, all call summaries and transcripts via chat)
• Monday priority planning (knowledge base into weekly focus list)

This list is growing fast. It’s also highly personal for now, but some of these skills will make sense for others too.

The takeaway

Start building your skills now. It takes time to figure out what you can automate and whether it makes sense. And fair warning: your first skills probably won’t work great. Descriptions need tweaking before the AI triggers them correctly, and you’ll learn over time what makes a good skill versus a messy one. That’s normal. The iteration is part of the process, same as the skills themselves improving over time.

But once you find something, it suddenly feels like you gained hours per week (to use for working more, jokes on you).

Some skills can be shared across teams. Others will be unique to how you work. But skills are how you move from “AI helps me occasionally” to “AI handles my workflow while I direct the outcomes.”

I think everyone will be there eventually.

If you’re curious where to start, the simplest skill is one that knows how you write. Collect the best pieces of your writing, give them to AI to analyze, ask it to make a reusable skill, and iterate on it every time you use it. Watch what happens to your drafts.

Where to use skills?

We’re launching skills in Desktop Commander too. Right now it’s a closed feature, but it works incredibly well for me. It has a skill builder (just ask the chat to build a skill for you), it automatically stores your skills and uses them whenever the chat context calls for it. If you’re interested in the closed beta, DM me and I’ll give you early access. Excited to roll this out to everyone very soon!

By the way, Cowork has skills now too. You can start by just asking the chat to help you create one and guide you through adding it. They also have built-in skills for working with MS Office files, and those are pretty neat.

In one way or another, they time to start is now!