What I'm Seeing#
I was talking with a business owner recently who told me he'd invested real money — hired someone — to build him a set of custom GPTs. If you're not familiar, a custom GPT is basically a version of ChatGPT you configure for a specific job: answer cu
stomer questions in your voice, help your team write emails a certain way, walk through a proposal in your format. When they work, they're great.
His worked. For a while.
But then things shifted — his services changed, his language evolved, the way he talked about his business changed — and the GPTs didn't. They got stale. He stopped using them. All that setup, sitting idle.
I've heard versions of this story more times than I can count. And it points to something real: the tools got ahead of the process. There was no easy way to keep them updated, no shared place to manage them, no clear ownership. So they drifted, and eventually got abandoned.
That's changing. Fast.
Right now, there's a lot of buzz around something called skills — and there's a good reason for it. But before I explain what they are, it's worth being clear about how they're different from what came before, because on the surface they can sound like the same thing.
A custom GPT is a destination. You build it, you navigate to it, it lives inside ChatGPT, and it does one specific job. It's a standalone tool — useful, but siloed.
A skill is a capability. It's a modular instruction set — plain text, just a file — that any AI agent can load and follow as part of a larger workflow. The difference matters because a skill isn't a chatbot you go visit. It's a procedure your AI carries with it. And you can stack them. An agent might load your brand voice skill, then your proposal-writing skill, then your quality-review skill — all in sequence, without a human handing off between steps. That's something a GPT can't do.
That's the shift. Skills aren't just a better GPT. They're organizational infrastructure. And the file format is now a cross-platform standard — the same skill runs in Claude, GitHub Copilot, Microsoft 365, and more. Write it once, use it anywhere.
A skill is basically a text file. Plain writing. It holds your methodology, your voice, your process. When an AI reads it, it knows how you work — no re-explaining, no starting over every session. And because it's just a file, you can update it yourself, share it with your team, and use it across tools.
This is why organizations are paying attention right now. Skills are a genuinely good answer to some persistent problems: Why does our AI output sound different depending on who wrote the prompt? Why does the new team member get different results than the senior one? Why do we keep getting generic content when we asked for something on-brand?
When everyone on your team is working from the same skill file — the same voice guide, the same process, the same quality bar — you get consistency. The AI becomes less of a wild card and more of a team member who actually knows your standards.
And here's where it gets even more interesting. As AI tools become more capable of working on their own — running tasks, managing workflows without someone watching every step — skills are what keep them on track. You're not hoping the AI figures out how you want things done. You're telling it, once, in a file, and it follows that every time. Autonomy with guardrails. That's the shift.
One more thing worth saying: skills don't learn on their own. The file stays exactly as you wrote it until you change it. That's intentional — you want deliberate rules, not instructions that drift. The good news is updating is easy. When something changes in your business, you update the file. Five minutes. Everything that uses the skill immediately reflects the change. That's exactly what was missing for the business owner I mentioned — a simple, shared place to keep things current.
The Anatomy of a Skill (the short version)#
A skill has two main parts: a description and the instructions.
The description is the most important part. It's short — a sentence or two — but it's what tells the AI when to use the skill. Anthropic (the company behind Claude) is clear about this: the description isn't a label for humans to read in a settings panel. It's the mechanism the AI uses to decide when your skill is relevant. Write it accordingly.
That's specific enough that the AI knows exactly when to pull it out — and exactly what to do when it does. Trigger phrases, document types, and the quality bar all named. Anthropic's own guidance says skills tend to under-trigger more than over-trigger, so lean toward being a little more specific rather than less.
The instructions are where everything else lives: your actual voice, your process, your standards, examples of what good looks like.
No code. No technical setup. Just clear writing with a sharp label.
Quick Win#
When I build skills for clients, I rarely start from scratch — I have the AI generate one from a session that already worked. Here's how you do the same:
The next time something really clicks — the AI nails your voice, the output is exactly right, the result is something you'd actually send or use — before you close that tab, ask:
"Based on everything we've discussed and the work we've done together, can you write me a skill file I can reuse? Include a specific description of when to use it, then the instructions that would help any AI recreate this quality of work for me."
What comes back is a ready-to-save starting point — not a vague note, an actual file.
Where you store it depends on your subscription. Most AI tools now have a built-in way to load your skill automatically — so you're not re-attaching a file every time. On some plans, you can store it inside a Project or as saved instructions that load at the start of every session. For teams, a shared folder works well in the meantime — one place everyone can find it, use it, and update it when things change.
The longer horizon here: as AI agents get more capable of running workflows on their own, skills become the files they reference automatically. No human needed to attach anything. The agent reads the skill, follows the rules, does the work. That's where this is heading — and building your skills now means you're ready when it gets there.
Try This#
Don't have a great session to pull from yet? Start here. Open Claude or ChatGPT and paste this:
"I want to create a reusable skill file for my business. Interview me about what I do, who I serve, how I like to communicate, and what good work looks like for me. Then write me a skill file with: a specific description of when to use it — including the kinds of tasks and phrases that should trigger it — and clear instructions any AI could follow to produce consistent, on-brand results."
You'll get a Q&A, then a draft. It won't be perfect — read it and adjust anything that doesn't sound like you — but it'll be 80% of the way there and a far better starting point than a blank page.
Save it somewhere. Name it clearly. Use it in your next session and notice how different it feels to start with the AI already knowing your standards.
Mahalo for reading — and for showing up curious, even when this stuff moves fast. That's the whole point of this newsletter: keep it real and keep it useful.
Want AI, Actually in your inbox every two weeks? Subscribe on Substack.
