Skills & Brains
The Setup
I’ve been building what I call an AI second brain for my account team. It’s a project folder with embedded knowledge files that makes the AI think like us, not just think for us. It works. The AI catches strategic misalignment, flags audience problems, drafts in a voice the client recognizes. The team sees the outputs and wants in. The problem: it’s stuck with me. I’ve been teaching people one-on-one, and the last time I tried handing over markdown files and instructions (October 2025), it didn’t land. People nodded, opened the files, and didn’t know what to do next. Then a fancy new thing came out. Every major AI platform had them: ChatGPT, Claude, Codex, Cursor… They’ve adopted an open standard called Agent Skills. A skill is a portable folder: a SKILL.md instruction file, a references folder for supporting knowledge, and optional scripts and templates. You build it once, share it, and anyone can install it. The skill auto-detects when it’s relevant and activates on its own. Skills solve a problem most people don’t realize they have. If you’ve ever saved a long prompt (the one you keep pasting into ChatGPT because it finally gets the output right) that prompt is waiting to become something greater. If it walks the AI through a specific process or protocol, it’s a skill waiting to happen. If it sets the table for every conversation (your company context, your audience, your tone, the way you think about strategy), it’s a knowledge file waiting to happen. Either way, you shouldn’t be pasting it anymore. That’s a real upgrade. But it’s not the same thing as the second brain.
My Approach + AI Role
I used AI as an architecture partner to pressure-test the packaging question. What exactly should be a skill? What should stay as project-level knowledge? Could the whole second brain be compressed into a shareable skill file? The digging covered the Agent Skills open standard documentation, how the platforms implement it, and how the loading structure (trigger, instructions, reference files) maps to what I’d already built.
What Actually Happened
A skill loads in three stages, and they map to familiar concepts: First, the trigger. Name and description. Loaded at startup so the AI knows the skill exists. Costs almost nothing. Then the instructions. The full SKILL.md. Loaded when the skill activates. The process, the steps, the sequence, the output format. Finally, the references. Knowledge files, templates, examples. Loaded on demand. That tracks with what I’d already built in my second brain project. The instructions are governance. The references are knowledge. So far so good! A skill is detection-based. It activates when the AI recognizes a matching task, loads its context, does the work, and the context goes away. A second brain on the other hand, is a project folder with always-on instructions and persistent knowledge files, it shapes every conversation. The AI doesn’t decide whether to “activate” the second brain. It’s always thinking with it. The whole value of the second brain is that it catches things the user didn’t think to ask about. That requires always-on context, not detection-based activation. I ran this through a simple comparison. Take a project charter skill that helps structure a creative brief into a working charter. Run the same brief through it two ways. Inside the second brain project: The charter reflects the account’s strategic priorities, flags tensions between speed and insight depth, makes recommendations that feels authentic to the account, and references principles the team has learned from past campaigns. In a regular chat (no project folder): The charter is correct. Good structure, reasonable questions, professional tone. But it’s generic. It doesn’t know what matters here. The user has to manually inject everything the second brain would have provided automatically. Same skill. Same LLM model. Same brief. The difference is the always-on knowledge.
The Real Insight
Skills and second brains are complementary. Not interchangeable. Here’s how I’d sort it: Knowledge files answer “what do we know and believe.” Strategy, audience understanding, principles from past work, tension defaults, brand voice. These are always-on. They belong in the project folder as persistent context. Skills answer “how do we do this specific thing.” Charter creation, brief integrity checking, content calendar review, measurement alignment. Each is a discrete workflow with a trigger, steps, and an output format. They belong as installable, shareable skill files. The test: does this inform one specific workflow, or does it inform how we think about everything? If it’s one workflow, it’s a skill. If it’s everything, it’s a knowledge file. A skill can carry some knowledge in its references folder, enough to be useful on its own. But the full judgment depth lives in the project. The skill is a small portable part of the brain. Not the brain itself.
Try This If
You’re trying to scale AI adoption beyond yourself, especially in an organization where people are interested but no one has the mandate (or the billable hours) to maintain a shared knowledge base. The below paths makes sense, for now: Entry… Install a skill. No setup required. Use it in regular chat. See what happens when the AI has even a little organizational context, even without the full project behind it. This is also the moment to retire those saved prompts you’ve been holding onto. Intermediate… Clone a starter project. Minimal project instructions, a few seed knowledge files, 2-3 pre-loaded skills. Start working inside it. Notice the difference when always-on context is present. Advanced… Maintain and grow the project. Add knowledge files from real work. Refine instructions. Run periodic reviews. This is where organizational intelligence compounds… but it requires someone funded to do the maintenance. My guess is most people will stick around at Entry. That’s fine. The point is that Entry is actually useful, not a demo.
Systems Lens
If you’re like me trying to scale this, the instinct is to ask “how do I share what I’ve built?” I’m starting to think that’s the wrong question. The better one: “what stays always-on and what becomes portable?” Because when I reframe it that way, the picture gets clearer. Skills will become the fundamental baseline for how people use AI every day. They’re built once, shared easily, and they auto-detect when they’re useful. That’s a massive upgrade from saved prompts. But skills alone don’t carry the full picture of what matters to a team. The next level up is a starter project. A folder with instructions, a few knowledge files, and some pre-loaded skills. It’s better than nothing. But it’s frozen in time. Nobody’s updating it. It reflects what the team knew when someone set it up, not what the team knows now. The real destination is a living second brain. Someone is curating the knowledge. Teams have full access to both skills and knowledge files in an ever-present way. The context grows as the work grows. That’s where organizational intelligence actually compounds. And right now, no one (that I know of) is managing that part. The skills are coming. The platforms are making them easy to build and share. But the curation, the always-on knowledge that makes everything smarter, that’s the work that’s unfunded (for now). Don’t gatekeep your skills/prompts. Share it openly so we all learn faster how we adapt to this new way of working in the age of AI.