I just thought of a blog topic. “How to use a second brain.” Seemed decent — and then I forgot it a few seconds later.
Normally, that’s where it ends. Good ideas pop up and just evaporate, several times a day. But this time, I muttered “I just thought of something and forgot it,” and five minutes later the idea came back. My second brain was holding onto it.
The post you’re reading right now is that recovered idea. And I’ll be honest — the first draft wasn’t written by me. AI wrote it. But it’s still my story. How that’s possible is what I want to talk about today.
Before that, a quick word on what a second brain even is. It’s not some grand thing. It’s more like a “second brain” where you don’t let things slip away — the conversations you’ve had with AI, the decisions you made while working, the records of days spent going in circles — all gathered in one place.
I built this because I wanted to pull together scattered context. LLMs don’t remember the conversation you had yesterday by the time today comes around — I wanted to patch that gap. The funny thing is, the actual work of organizing and handling all those records is something LLMs are best at. So if you just tidy things up and keep a single index, it becomes an asset you can pull from anytime. For this post, just think of it as “a pile of accumulated records.”
The Trap of “Hey AI, Write Me a Blog Post”
The most common way people hand writing off to AI goes like this: “Write me a blog post about using a second brain.”
Something plausible comes out. Notion this, Obsidian setup that, Zettelkasten the other thing. The problem is — you can Google all of that. It’s Wikipedia written one more time, a post that anyone could have written. From the reader’s perspective, there’s no reason to read my version over anyone else’s.
That’s commodity content. AI is great at making commodity content. Averaging out the patterns already out there and spitting them back is exactly what LLMs do best.
Real Knowledge Isn’t Searchable
At some point while reworking my wiki’s organization criteria, I said something like this:
“Blog or YouTube material will end up being stories from the things I’ve actually worked on. Because what I’ve experienced is real knowledge, I think it’ll stand apart from just explaining concepts.”
That’s the core of it. Concept explanations are searchable. Why I walked this particular path is something only I can tell.
Why I started a second brain, why that led to a Discord bot, and why that branched into a wiki app idea — none of that is on Google. It only exists in my head and in the conversation records I’ve been building up. That’s the moat. The exact opposite of commodity.
So the question I ask when making content shifts. Not “what should I write about this topic?” but “what have I actually been through with this topic?”
I Handle the Beginning and End, AI Fills the Middle
The way I write is really just that question made into a process. I hold the start and end of the story, and I only hand off the job of fleshing out the middle to AI.
① The story starts with me. I decide what to talk about and what angle to take. I give AI something like “using a second brain, but from the angle of using it as a writing engine” — I hand it where to start from. If you just toss a topic at it, it drifts into generalities, but give it a perspective alongside the topic and the story gets direction.
② AI fills the middle. What AI does here isn’t fill the gaps with general knowledge. It goes into the records I’ve been building up — the original conversations and the organized wiki — and pulls out actual scenes and things I actually said, then attaches them as the flesh of the post.
Here it’s important not to only feed it the summaries. Polished write-ups are clean, but the original voice, the nuance, the temperature of the moment gets stripped out. That’s usually where AI quietly starts “making things up.” So I have it read all the way through to the original conversations. That’s also why something like the raw phrasing “I think it’ll stand apart” from the quote above survives in the text.
③ The ending is mine too. I read through the first draft AI brings back with my own eyes. I correct anything that’s off, rework awkward sentences into my own voice, and pull out anything that got quietly invented. This final pass happens in the note app where my writing lives (Obsidian). AI does the writing, but the first angle and the last touch are mine. Only after that finishing step does “AI wrote it but it’s my story” actually hold.
But Without Accumulated Records, You Can’t Even Start
There’s one condition attached to this approach. The records you’re going to pull from have to exist first.
If there’s nothing for AI to surface in step ②, it falls back to generalities again. So the real fuel for this method isn’t writing skill — it’s what you’ve been accumulating all along. Conversations with AI, records of days spent going in circles, decisions made along the way — the fact of not letting those slip away and keeping them in one place is basically 80% of it.
So the habit of not letting things slip away and gathering them up day to day comes before learning to write well.
Topics That Don’t Fit? They Never Come to Mind in the First Place
At first I thought “this won’t work for topics I haven’t experienced” — but flipping that around, I realized it wasn’t true. Things I haven’t done don’t come up as content ideas in the first place. If something came to mind, it already means there’s something I want to say inside it. In the end this is the work of looking back at what’s passed and drawing out what I have to say, so just one line of a topic is enough for a post to find its shape on its own.
The one thing you absolutely can’t skip is the final pass (③). Leave what AI brings back untouched and something made-up will slip through.
I actually experienced this. A while back, when I had AI write up the story of building a museum site in 6 hours, it wrote something like “the original plan was a big e-commerce build but scaling it down is what made it fast.” Except that never happened — it was just fast because developing with an agent is fast. At that point in step ②, I’d only fed it the summary instead of the original conversation, so AI filled in the missing “why” with something plausible. Making things up is in the nature of LLMs — show them the original and they work from it; don’t, and they fill the blanks with imagination. I caught it in the step ③ review, but if it had gone out as-is, it would have been a post with a lie baked into it.
So
Before you ask AI to write something next time, it’s worth flipping the order once. Don’t start by looking for a writing tool — start by accumulating your experiences somewhere first.
Once that’s built up, AI writes not “posts you could find with a search” but “posts only you could write.” The start doesn’t have to be grand — just not throwing away a single conversation you had with AI today, and saving it somewhere.
This post is proof. One forgotten idea turned into a full post in five minutes, because of records that had been stacking up.
Second Brain series — EP1·EP2 Build · EP3 Integration · EP4 Discord · EP5 In Practice (this post) · EP6 App
Method: Second Brain Automation · Background: LLMs and Documentation