Your AI training worked. Now what?
We've trained a lot of teams on how to effectively use generative AI since 2022. The sessions go well. People leave excited and often saving 4–12 hours per week. They start using ChatGPT, Claude, or Copilot in their actual work. Some of them get really good at it.
Then something happens that most people don't talk about.
A few weeks in, the people who got the most out of training are spending a surprising amount of their AI time doing the same thing over and over: re-explaining context. While we teach people about custom instructions, projects, and custom GPTs for everyday quick chats, conversations start at zero. And even when working in projects with specific instructions, they don't have the understanding of instructions and other projects. They're retyping brand guidelines, re-describing the approval process, and pasting in the same setup prompts from a Google Doc or similar that they keep open in different tabs.
It works, but it becomes a mess.
We've also noticed this with our own team at Mostly Serious. We have people who have built genuinely useful AI workflows that automate some portion of their work. They've written powerful prompts, useful instructions, and real quality guardrails that keep the AI moving in the right direction. Unfortunately, most of this lives in their heads or scattered across their individual uses of AI tools. And even in their own use cases, when they start a new conversation, they have to reconstruct the AI's understanding of what they're setting out to accomplish. When somebody else on the team wants to do the same thing, they're doing this individually too.
The knowledge was there. It just wasn't shared and portable.
What we started doing differently
A few months ago, I started packaging my best prompts and workflows into something called Skills. A skill, originally created by Anthropic, is a set of instructions, written in plain language, that teaches the AI tool how to handle a specific type of work. You write it once. The AI follows it every time.
It's not code or an integration, but rather a markdown file with clear directions. A lot like the custom instructions we teach people to create, but portable and shareable.
I built one that matches my personal writing style, trained on years of my own journals, tweets, and articles. Another one that pressure-tests my arguments before I publish them. One that applies our Mostly Serious brand guidelines to anything the AI produces. A deep research skill that turns Claude into a proper research partner instead of a summarizer.
After that, I stopped re-explaining myself. The AI already knew what I expected because I'd written it down once in a place it could find. And maybe most importantly, I'm able to use these skills in any conversation, when I need them.
What changes for teams
Skills are useful for individuals. But they're even more interesting for organizations.
When your best AI user packages what they've figured out into a skill, anyone on the team can use it. The marketing coordinator who joined last month gets the same quality output as the person who spent three months refining their prompts. The finance team runs the same quarterly report without re-explaining the format. An operations lead's vendor evaluation process stops living in their head and becomes available to everyone.
At Mostly Serious, we call these House Skills. A central hub where anyone on the team can share and refine the tools that help all of us move faster.
Looking ahead, these skills will also be used by our AI agents. When a project is being created and one of our agents needs a skill — let's say front-end design — it can grab that skill and keep working just like a human would. This is important because any work we do to mature our AI use for people should also cross over to maturing our AI use for agents. That's where the value is going to be found as these tools continue to mature.
The compounding part
Skills can change how AI training compounds inside an organization.
Without skills, AI training is a one-time event. People learn, some of them stick with it, and the gap between your best users and everyone else slowly widens.
With skills, training becomes a starting point. The things your team discovers get captured and shared. The gap narrows. The whole team moves up instead of the few people who happened to take to it naturally. Because the practical use cases aren't siloed.
If you've invested in AI training and haven't started building skills yet, that's where I'd look next.
We put together a guide
Like many things in the AI space, the big companies love to create new buzzwords to describe how things work. We wrote a playbook that explains what Skills are, how they work, and what they look like in practice. It's all a lot more straightforward than you probably expect.
It covers the three main types of skills we've seen work well, gives real examples across departments, and walks through how we build them for clients.
Written for real teams without the buzzwords or hype.