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.
Press enter or click to view image in full size
Skills are enabled automatically based on what you prompt your LLM.
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.
> This is what AI skills unlocked for me, and I want to walk through how.
That one sticks out.
I hate having to recalibrate how I read comments on websites. We are basically in the spam days of e-mail without any filter in sight — gasp! an em-dash! totally sus… — the cost of writing an extensive comment in faultless prose has fallen to nearly nothing, so it gets posted. Just like spam the reason for doing so seems to be for minimal effect, to move a discussion in some direction by minute degrees, or simply for the attention — perhaps to build up a credible-ish account?
Faultless in terms of grammar and orthography, but very verbose, horribly hard to focus on, full of emptiness and terribly boring, each being way more faulty than a few typos and language mistakes. Maybe I'll just start flagging these comments and HN posts more systematically.
It feels human to me, but LinkedIn-ish. The "Press enter or click to view image in full size" suggests this was cut and pasted from a source blog or something.
My workday is fully automated now using this tool that I’ll sell you…
Unfortunately obviously written with AI - overly focused on everything being 'meaningful', 'real', 'actual', etc.
I've never felt like automating my days, be it work, hobby or whatever. Programming sure, I do use AI under supervision
So far 1/6 comments on this post and the post itself are AI slop
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.
Press enter or click to view image in full size
Skills are enabled automatically based on what you prompt your LLM. 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.
I can feel a lot of AI tropes in this writing.
I get a few lines in, recognize the "explain this as if you are an excited highschooler who can't stop talking" style, and lose focus.
> This is what AI skills unlocked for me, and I want to walk through how.
That one sticks out.
I hate having to recalibrate how I read comments on websites. We are basically in the spam days of e-mail without any filter in sight — gasp! an em-dash! totally sus… — the cost of writing an extensive comment in faultless prose has fallen to nearly nothing, so it gets posted. Just like spam the reason for doing so seems to be for minimal effect, to move a discussion in some direction by minute degrees, or simply for the attention — perhaps to build up a credible-ish account?
Faultless in terms of grammar and orthography, but very verbose, horribly hard to focus on, full of emptiness and terribly boring, each being way more faulty than a few typos and language mistakes. Maybe I'll just start flagging these comments and HN posts more systematically.
Have been intentionally not fixing typos in anything I post as a shibboleth, hoping the slop producers dont catch on.
It feels human to me, but LinkedIn-ish. The "Press enter or click to view image in full size" suggests this was cut and pasted from a source blog or something.
I'm sorry you were able to read that far so as to notice this gem. Ah ah.
It's a copy paste of the medium article. That's not what HN comments are supposed to be.
It suddenly feels like the future is being built by the most careless and least detail-oriented people. How did we let things become this stupid?
"Move fast and break things?"
ai;dr
tldr: an ad for an ai wrapper
Yep, I wish there was something to filter these glorified advertisements out