The Skill Stack
Most people use AI backwards. They ask it to generate when they should be asking it to transform. Here's the difference.
Most people using AI to write are doing it wrong.
They type "write me a blog post about productivity" and get back 800 words of perfectly structured nothing. It reads fine. It says nothing. The verbal equivalent of elevator music.
Then they wonder why AI-generated content all sounds the same.
The mistake is treating AI as a generator when it's actually a transformer. That distinction matters more than anything else you'll learn about these tools.
Generation means asking the machine to conjure something from nothing. You get what's in the training data—which is everything, which means nothing in particular.
Transformation means feeding it your raw material and asking it to refine. Your voice memo becomes a draft. Your transcript becomes an essay. Your scattered notes become a coherent argument. The AI doesn't replace your thinking. It extends it.
Garbage in, garbage out. Everyone knows that.
But flip it: treasure in, treasure out.
The Fairy Godmother worked her magic because Cinderella had something worth transforming. A heart of gold, even in rags. The magic revealed what was already there.
Same principle. Your source material is your moat. Your transcripts, your notes, your half-formed ideas captured at 2am—that's the ore. AI is the refinery.
I learned this by accident.
In 2023 I was polishing podcast transcripts, the kind of tedious work that makes you question your life choices. Hours of "um" and "you know" and sentences that started three times before finding their footing.
Then I highlighted a paragraph in Notion and clicked "Improve writing."
Three seconds later: the same ideas, tighter. The speaker's voice intact. Just... better.
I did it again. And again. An hour of editing collapsed into minutes.
That was the moment I understood. AI wasn't going to write for me. It was going to write with me. The raw material was still mine. The judgment was still mine. But the friction between rough and polished had nearly disappeared.
The problem with prompts is they don't remember anything.
Every conversation starts from zero. You explain your voice, your goals, your preferences—then do it again next session. And again. The AI has amnesia. You're the only one keeping track.
The second brain movement tried to solve this. Tiago Forte, August Bradley, the whole PKM ecosystem. Elaborate systems for capturing and organizing knowledge. Beautiful in theory. Maintenance hell in practice. Endless copying and pasting between apps that don't talk to each other.
Skills fix this.
A skill is a markdown file that makes your AI smarter. That's it. No API. No code. Just text that teaches.
You write down how you work—your voice, your patterns, your preferences—and the AI loads it automatically. Your context persists. Your standards compound. Every session starts where the last one ended.
.claude/
├── CLAUDE.md # Who you are, how you work
└── skills/
└── voice-matching/
└── SKILL.md # Instructions + examples
The technical term is progressive disclosure. The AI loads what it needs, when it needs it. But the effect is simpler than that: you stop repeating yourself.
Most skill collections miss the point entirely.
Browse GitHub for "awesome Claude skills" and you'll find repos full of generic instructions. "Write in a professional tone." "Be concise and clear." "Consider the audience."
This is useless. It's already in the training data. You're not teaching the model anything it doesn't know.
A good skill contains knowledge the model can't have without you:
Your actual writing samples, annotated.
Your specific anti-patterns—the phrases you never use.
Your workflow quirks, your client requirements, your house style.
The difference between a mediocre skill and a powerful one is the difference between a job description and an apprenticeship. One tells you what to do. The other shows you how someone actually does it.
I run content for an education company. Newsletters, podcasts, social, course materials. The whole operation runs through Claude.
What used to require a team—copywriter, editor, podcast producer, social manager—now runs through a stack of skills I built over months.
Not because AI replaced those roles. Because skills let me bring the judgment of each role to bear on everything I touch.
Voice Matching catches when drafts drift toward generic AI prose. Anti-AI Writing kills the telltale patterns—the correlative constructions, the throat-clearing, the hedge words. Transcript Polisher turns raw audio into readable text without losing the speaker's quirks.
None of this is magic. It's documented knowledge. The power comes from stacking.
The models are good enough. They've been good enough for a while.
The bottleneck now is context. Your ability to load the right knowledge at the right time. Your ability to build skills that compound. Your ability to create a system that learns because you taught it what to learn.
This is what I mean by context engineering. It's the skill behind the skills.
Templates are training wheels you never take off. Skills are the bicycle.
In 1997, Garry Kasparov lost to Deep Blue. Chess was "solved." Humans would never beat machines again.
What happened next is more interesting.
A new format emerged: freestyle chess. Human-machine teams competing against each other. And a strange pattern appeared. The best teams weren't the best humans or the best machines. They were average players with above-average skill at collaborating with their computers.
The combination beat everything.
That's where we are with writing. Pure human or pure AI, you lose. Human with skills—with context engineered to amplify your judgment—you win.
The gap is widening. Not between people who use AI and people who don't. Between people who engineer their context and people who type into a blank chat window.
There's a version of this that's dystopian. Black-box automation. Someone else's workflow imposed on your problems. You don't understand what's happening. You can't modify it. You're dependent on systems you didn't build.
Skills are the opposite.
You build them. You modify them. You understand exactly what's happening. The knowledge lives in markdown files you can read. When something breaks, you fix it. When something improves, you keep it.
This matters more than convenience. It's the difference between using a tool and being used by one.
Today, a single writer with a well-built skill stack can do the work of an entire production team.
I don't mean this theoretically. I mean I watched my own job transform. The tedious parts evaporated. What remained was judgment, taste, the work that actually matters.
The person who masters context engineering makes their old job obsolete. Not in the dystopian sense. In the liberation sense.
That's what this newsletter is about. Each week, one skill. Portable. Documented. Yours to modify.
The model is smart enough. The question is whether you'll catch up.