Every AI tech bro on the internet is currently building a second brain and would love to tell you about it at length on their podcast or YouTube channel. Annoyingly, they have a point. There are several good reasons to build one, and my current favorite is being able to tell Claude to piss off.
Lately, Claude and I had been disagreeing about who was in charge. I’d ask for something specific, and it would decide it knew better. Having access to my business files seems to have given it the impression we were partners. We aren’t splitting the expenses, so I’m unclear on how we got there.
I kept having to explain to a computer that I’d already thought about what I wanted before I asked for it. That’s an irritating conversation to have with a person, and having it with software I voluntarily invited into my life AND PAY FOR adds a special layer of embarrassment. I have enough people telling me what to do without a bot piling on.
So I switched to ChatGPT, pointed it at my second brain, and kept working. Mine is split between Obsidian and StorytellerOS, and the new tool could read all of it. I’d already written down how my business works and how I want things done, so I got to act on my irritation right away, without first calculating how long it would take to train another AI from scratch.
I’m not married to any of these tools
I’ll say it again. All of these companies are in an arms race. They are going to keep releasing new models, and I’m going to keep using whatever works for me at the moment. I don’t plan on a long-term commitment to a product that might behave differently after its next update.
Some of you have read a lot of Claude tutorials from me, so this email may feel like I showed up to the family reunion with a different date. Those tutorials were based on what worked at the time. I’ll keep telling you when something works and when I’ve moved on to something better. I’m comfortable with that level of inconsistency in my relationship with a software company.
Let’s just say my mantra of late is… why choose? (Insert reverse harem jokes here.)
The competition has produced some extraordinary capabilities, and I’m also watching for regressions, which is the polite technical term for an update making something worse. A model can get better at difficult work and somehow get worse at an instruction it used to follow without complaint. A skill you spent time getting right may need revising because the new version interprets it differently.
The release announcements focus on the improvements, naturally. You’ll usually discover the rest while trying to finish something you’d told yourself would take ten minutes.
Start with a folder
“Build a second brain” can sound like an invitation to spend a weekend arranging folders when you should be doing the work inside them. I enjoy an unnecessarily elaborate system as much as anyone, so I’m probably the wrong person to supervise you near a blank dashboard.
What matters is getting the information out of your head and out of any one AI app, in a form you can use again. In my setup, that’s written context about my business and instructions for how I want work handled.
I keep mine in two places. Obsidian holds my notes, decisions, and AI instructions as plain text files on my own computer, which I can open with or without any app at all. It’s exceptional for this because notes link to each other and an AI can follow those connections, though I’ll admit it takes more setup than most people want on a Tuesday night. StorytellerOS holds the author side of the business: series, manuscripts, retailer links, pen names, newsletters, and social posts. I built it because my author business was scattered across more tabs than I care to admit. Whichever AI I’m using that week can connect to both.
Plenty of people do this in Notion, and plenty more just use files and folders on their computer. Any of those work if you can open your documents without the AI.
You can start with a folder. The name of the folder is not a decision that requires a research phase.
What to put in it
Some of it is straightforward. For each series, list the books in reading order with links to where readers can buy them. (Mine sit together in StorytellerOS, so I’m not hunting down the right buy link at midnight.) Add enough explanation that an assistant knows which book you’re promoting. You probably don’t want it inviting new readers to start with book four, right after someone they were supposed to get attached to in book one has been murdered.
If you write under more than one pen name, make the separation explicit. Your sweet romance readers and your monster romance readers may have different expectations for a newsletter surprise, and leaving the AI to guess which audience you mean is an unnecessarily exciting way to spend release day. I keep each pen name walled off in its own space, and I’d recommend some version of that wherever you keep yours.
The harder part is the stuff you’ve stopped noticing you know. You’ve decided how often you’re comfortable emailing readers. You remember why you quit a particular promotion. You probably have a perfectly good reason for sending people to one retailer in certain situations, and another for skipping a popular tactic that doesn’t fit your books. Write those decisions down with enough context to be useful, or you’ll keep getting sensible suggestions you already considered and turned down, and you’ll spend the time you meant to save explaining yourself to a chatbot.
Show it how you write
“Friendly and conversational” will get you approximately nowhere. Give the AI examples of things you’ve actually written, tell it what sounds off in its drafts, and save the corrections you keep having to make. The ones that make you want to delete the paragraph and do it yourself are the most important to keep.
An assistant needs to see how you explain an idea, including where you stop explaining because your readers are intelligent adults who can follow a thought on their own.
(I have feelings about this. You may have noticed.)
My voice samples and past newsletters live in the same place I write and send the new ones, which means the assistant drafting this week’s email is working from the real thing. The more detail you capture, the less you rely on an app remembering a conversation correctly. You also end up with a record you can read. If an instruction is wrong, you can fix it yourself without trying to work out what the bot thinks it knows about you.
Memory and skills
Memory is the information about your business and the decisions you’ve made. A skill is a set of instructions for doing one particular job, like preparing your weekly newsletter.
“Help with my newsletter” leaves an impressive amount of room for interpretation. A useful skill tells the assistant who the audience is, where the voice guide lives, how long the finished draft should be, where to save it, and what still needs your review.
It can be rough on the first pass. You’ll find out what’s missing when you use it, and that’s a lot more helpful than spending three days imagining every possible exception.
Plan on some upkeep. A new model might keep misreading one of your instructions, or a tool might change how it connects to another app, and you’ll need to revise that part. Keep the current version with the rest of your materials. If I’ve spent an afternoon working out how to get a good result, I want to carry that work into whatever tool I use next.
I’m teaching a three-hour session on building your second brain at Author Nation. We’ll get the information you keep explaining to AI into a system you can take with you when you switch tools. [The workshop details are here.]
Keep a copy you can actually open
An app assuring you it will remember something is easy to mistake for a filing system.
You should be able to open your documents without opening the AI app, and they should make sense when you read them. Keep a backup somewhere else, too, so an unavailable service doesn’t take your only copy with it. I follow my own advice here. My Obsidian files sit on my computer and get backed up separately, and I export from StorytellerOS too, because I’d feel ridiculous writing this newsletter otherwise.
If your instructions only exist inside a project in one AI app, find out how to export them while the app is working and you still have access. Then read the export. A huge archive of conversations might have what you need buried in it somewhere, and I don’t want to spend the morning of a launch looking for it.
Your current AI can help turn those conversations into usable documents. Ask it to pull together what you’ve decided about a particular task, and then check its work. It may leave out a correction or include an idea you talked about and never agreed to, and you’ll want to catch that before it becomes instructions for the next assistant.
Reading it over also has a way of showing you what you never actually wrote down, or where your own instructions contradict each other. I would love to tell you every communication problem starts with the bot. Unfortunately, I have met us.
Why this matters for your business
Keeping the only copy of your memory and skills inside one AI app is a recipe for disaster. We’re handing these services more of our daily work, so their business decisions can turn into our business problems quickly.
Prices can go up. You might still like the product, but liking it doesn’t mean you have to pay whatever it asks. If you can move your information, you have a real choice when you decide whether the new price makes sense. If you can’t, you’re comparing the new monthly bill to the time it would take to rebuild everything you’ve taught the app, and that’s a miserable little calculation after months of getting things the way you want them.
Terms can change, too. Something you were fine with at signup might not be acceptable anymore, and I want to make that call based on the terms without worrying that leaving will cost me my own working instructions.
Services also go down. The company may be doing everything it can to fix an outage, and your release email still has to go out. If the only copy of your process is in that AI app, you’re stuck trying to remember which links go where and rebuilding the instructions that kept your drafts from sounding like a motivational speaker wandered into your inbox. My release emails and social posts go out from StorytellerOS, so when one AI is having a bad day, I connect a different one and keep going while the first one sorts out its emergency.
Switching still takes some work
A skill may need adjusting for the new app. Connections to your other services might have to be set up again, and you’ll need to check the results, because a new assistant reading your instructions won’t necessarily follow them perfectly. You’ll have the source material and the decisions you already made, though, and that’s a much more manageable job than trying to remember everything you’ve ever told an AI.
You can use the same material with more than one tool if you like doing different jobs in different apps, and you won’t have to maintain several competing biographies of yourself. It’s useful for humans, too. If you hire an assistant someday, you can give them a login to where everything already lives, which beats handing over your inbox and a hopeful plan to remember the important bits.
If you only do one thing
If this has you mentally adding “document my entire existence” to a list you already resent, start with the recurring job where you’re most tired of correcting the AI.
Ask your AI to draft instructions for that job using the context it has.
Read the draft as if you were handing it to someone on their first day. Add the decisions it missed and remove anything it made up, before a confident guess turns into company policy.
Save the corrected version outside the AI app, along with an example of a result you were happy with. (Obsidian, StorytellerOS, Notion, or a plain folder all work. Pick the one you’ll actually open again.)
Give it to a different AI and ask it to do the same task.
You’ll see what carries over and what still needs explaining. You might decide your original tool does a better job, and that’s good information to have. You’ll still own the document.
Where to get help
There are new free classes in the Author Automations Hub for your burning AI questions. If the idea makes sense and you get stuck trying to make it work with your own files, bring that with you. The part where the instructions seem reasonable and the software stays deeply uncooperative is a very familiar activity around here.
We’re also planning an AI Rodeo in San Antonio next year, and I’ll share more as plans come together. I’m looking forward to getting a room full of authors working on this together. It’s reassuring to watch somebody else’s supposedly helpful assistant need a little adult supervision.