Explorer
Home Blog Contact Claude SDD

I'm Letting Claude Manage My Life (Sort Of)

Jan 13, 2026

So I’ve been on this journey of figuring out how to best work with AI coding assistants for a while now. I’ve written about my vibe coding dead end, my attempts at structure, and my meta-repo workflow. But recently I’ve been experimenting with something a bit more… ambitious. What if Claude wasn’t just helping me code, but helping me manage my entire life?

Before you start yelling about that Wall-E future, let me explain.

The Second Brain(s) Situation

I’ve written before about getting deep into the “Second Brain” methodology. What I haven’t mentioned is that I actually have two second brains. Yes, I know. Classic ADHD trap, overoptimizing your system as a way to procrastinate from doing actual work.

It started a few years ago when I fell down the Ali Abdaal YouTube rabbit hole and became convinced that Notion was going to be the magic bullet that finally cured my ADHD. Spoiler: it wasn’t. Don’t get me wrong, it helps, but building elaborate Notion systems can become its own form of procrastination. Ask me how I know.

I’ve taken a lot of inspiration from David Allen’s GTD and Tiago Forte’s Building a Second Brain concepts over the years. But I’ve heavily modified both for what actually works for me. The full PARA system never really clicked, for example - I found myself fighting the categories more than using them. Same with strict GTD contexts. I’ve cherry-picked the parts that help and ignored the rest. My system is a Frankenstein’s monster of productivity methodologies, and I’ve made peace with that.

But I stuck with Notion because it genuinely excels at certain things. The database functionality is fantastic. Linking information together, building quasi-dashboards, creating relational views - it’s all pretty seamless. I use it for:

  • Habit and task tracking - Daily habits, recurring tasks, the whole GTD-lite thing
  • Job hunting - A database of companies I’m interested in, jobs I’ve applied to, interview stages
  • Shared social calendar - My partner and I track venues we’ve been to, events we might want to attend, things like that
  • Bookmarks/Read Later - Quick clips from the web

The problem? Notion is someone else’s computer. All that data lives on their servers, in their proprietary format. They have export features, sure, but have you ever tried to actually use a Notion export? It’s rough. And I kept having this nagging thought: what if I have a decade of private journal entries in here and then, poof, Notion goes away? Or gets acquired? Or decides to triple their prices?

That’s where Obsidian comes in. Local-first, plain markdown files that I own forever. If Obsidian the company disappeared tomorrow, I’d still have a folder full of .md files that any text editor can open. For anything I want to keep long-term - journals, learning notes, personal reflections - that peace of mind matters.

So now I run both. The litmus test I use: “Will I care about this in five years?” If yes, it goes in Obsidian. If it’s more ephemeral, collaborative, or database-heavy, Notion. Is it a little annoying to juggle two systems? Yeah. But each does something the other can’t, and the collaboration features alone (sharing that social calendar with my partner) keep Notion in the rotation.

You can hack together database-like functionality in Obsidian. For years, the community Dataview plugin was the answer - write queries against your notes’ YAML frontmatter and render tables. It works, but it’s code-heavy and read-only. Then Obsidian finally released Bases as an official core plugin. It turns your notes into editable, filterable database views based on their properties. It’s a big step forward, but it’s still table-view only - no kanban boards, no gallery views, no calendar views like Notion offers. And it doesn’t support inline properties or images in the table view yet. It’s getting there, but it’s not Notion-level seamless. Trade-offs everywhere.

Task management is similar - Obsidian has plugins like Tasks that can handle it, and I might eventually consolidate my tasks into Obsidian too. The biggest barrier is mobile. Notion’s mobile app is polished and fast. Obsidian’s mobile experience is… functional, but clunky enough that I don’t want to rely on it for quick task capture throughout the day. I’ve also used TickTick in the past - fallen off a bit lately, but I’m curious about building some kind of Obsidian ↔ TickTick sync flow. Best of both worlds, maybe. We’ll see.

The Problem

So I’ve got Notion, Obsidian, my ideas/ meta-repo for project planning, and my actual code repositories. Four systems, zero integration. My Obsidian vault didn’t know about my coding projects. My coding assistant didn’t know about my personal goals or what I studied yesterday. Notion was off in its own world entirely. Context was scattered everywhere.

The ADHD brain in me was not thriving.

The Experiment

I created a new repo called my-life (very on the nose, I know). It sits above everything else and gives Claude unified context across:

  • SecondBrain/ - My Obsidian vault (journals, notes, learning)
  • ideas/ - Project planning meta-repo (specs, issues, briefs)
  • .claude/ - Skills, memories, and learning session data

“But wait,” you say, “what about Notion?” Yeah, that’s still mostly separate. Notion does have an MCP (Model Context Protocol) server, so theoretically Claude could talk to it. But here’s the thing - Obsidian’s killer feature for AI integration is that it’s just markdown files. Claude can read and write them directly, no API calls, no authentication dance, no rate limits. It’s so much easier to work with.

Notion might get integrated eventually, but for now, the stuff that benefits most from AI assistance - my journals, my learning notes, my long-term knowledge - all lives in Obsidian anyway. The ephemeral Notion stuff (task lists, habit tracking) doesn’t really need Claude’s help.

The key insight is that Claude Code can read from all of these simultaneously. So when I ask it to help me plan my week, it can see what I worked on yesterday (from my journal), what issues are open in my projects (from ideas/), and what I’ve been studying (from my learning sessions). One assistant with full context.

Skills Pay The Bills

I’ve been building some Claude Skills to help me with this. Things like:

  • /good-morning - Reviews yesterday’s journal (and makes sure I actually wrote one), checks what’s due for studying or review, helps set up what to focus on for the day
  • /daily-review - Pulls my GitHub commits, asks about what I did, updates my journal
  • /start-session - Begins a structured learning session on a topic
  • /youtube-catchup - Summarizes new videos from channels I follow (so I can decide what’s actually worth watching)

These aren’t magic. They’re just well-structured prompts that tell Claude exactly what to do, what files to read, and what format to output. But they save me a ton of mental overhead.

The Memory System

This is the part I’m most excited about. Claude doesn’t actually remember anything between sessions. Every conversation starts fresh. But what if we could fake it?

I’ve set up a simple memory system where Claude proactively captures things it learns about me during our conversations - preferences, corrections, context, workflow insights. These get saved as JSON files that Claude reads at the start of each session. It’s like leaving notes for your future self, except the notes are for your future AI assistant.

The key word there is “proactively.” I explicitly told Claude to just do this without asking me. If I say “actually, I prefer X over Y,” it should just capture that as a memory. No friction, no prompts, no “would you like me to remember this?” Just do it.

The Learning Workflow

This one’s probably my favorite part of the whole setup. I wrote a while back about using ChatGPT as a personal tutor, and it worked surprisingly well - for a while. But I called out a concern in that post: “Eventually I will hit the limits of its context memory.” And yeah, that happened. The longer I used it, the worse it got at remembering older material. Quizzes became less relevant, callbacks to previous sessions got fuzzy. The chat-based approach just doesn’t scale.

The fix? Get everything out of the chat and into files. Session logs, learning plans, progress tracking - all stored as markdown in my repo. Now Claude can read my entire learning history at the start of each session. No context window limits because the memory is external.

I was already keeping all my learning notes in Obsidian - course notes, book notes, technical concepts, stuff I wanted to retain long-term. Anytime I “learned” anything. Each note has frontmatter with lastReviewedDate and reviewFrequency properties, and I’d use Dataview queries to surface what was due for review. I also use the Spaced Repetition plugin to embed flashcards directly in my notes using a simple question :: answer syntax.

It worked, but it was manual. I had to remember to check my review queue. I had to write the flashcards myself. And “reviewing” often meant just re-reading notes, which is… not actually how memory works.

Now Claude handles most of this:

  • /start-session [topic] - Claude teaches me a topic conversationally. It starts with a retrieval warm-up (“what do you remember about X?”), identifies gaps, then fills them in. It logs everything to session files so we can pick up where we left off. I still manually take notes as we do this to help with retention.
  • /review-session [topic] - Pure retrieval practice. Claude quizzes me on material I’ve studied, grades my answers, and tracks what I’m struggling with.
  • /flashcards [topic] - Generates flashcards in the Spaced Repetition plugin format from my notes or from a topic we just covered.

The magic is that Claude can actually teach, not just present information. It asks questions, waits for me to attempt an answer, then corrects misconceptions. That active recall is way more effective than passive re-reading. And because it’s tracking sessions over time, it knows what I’ve covered, what I got wrong, and what’s due for review.

Still early days, but this might be the most valuable part of the whole experiment.

The YouTube Problem

I watch TOO much YouTube. But I kept running into the same thing: I’d search YouTube for something and see that half the results were videos I’d already watched. I wasn’t retaining any of it. So a while ago I started taking video notes in Obsidian. That’s great if, again, a bit manual. I’d watch a video, take notes, and then I’d have a record of what I’d learned.

But the bigger problem is volume. I’m subscribed to way too many channels. Tech channels, productivity channels, Atlassian channels, AI channels. There’s no way to actually watch everything, so I’d either fall hopelessly behind or waste hours skimming videos that turned out to be fluff.

So I’m piloting a new workflow: /youtube-catchup. Claude pulls the latest videos from a list of channels I care about, fetches the transcripts, and generates summaries with a “Why Watch?” recommendation. The summaries go into my Obsidian vault as video notes, tagged and searchable. Now instead of watching everything, I skim the summaries and only actually watch the ones that seem genuinely worth my time.

It also extracts “discoveries” - products, tools, frameworks mentioned in videos that might be worth looking into. Those get created as stub notes in my inbox for later triage.

Just started this today, so jury’s still out. But the promise is: stay informed without drowning in content. We’ll see if it holds up.

What’s Working

  • Unified context is powerful. Being able to say “based on what I’ve been working on this week, what should I focus on today?” and getting an answer that actually considers my journal, my projects, and my learning goals is genuinely useful.
  • Skills reduce friction. Instead of explaining what I want every time, I just run /daily-review and Claude knows the drill.
  • The memory system is surprisingly effective. Even with simple JSON files, Claude can maintain continuity across sessions in a way that feels much more personal.

What’s Still Rough

  • It’s a lot of setup. Getting all these skills and workflows dialed in takes time. This is very much a “sharpen the axe” situation.
  • Context window limits are real. Claude can’t actually read my entire vault at once. I have to be strategic about what it loads (maybe use a RAG in the future)?
  • I’m essentially building a custom system. This isn’t something I can just hand to someone else. It’s deeply personalized to my specific workflow.

Is This Overkill?

Probably? But here’s the thing - I have ADHD. Systems and structure aren’t optional for me, they’re deeply necessary. And if I can offload some of the cognitive overhead of maintaining those systems to an AI assistant, that’s a win.

Plus, it’s kind of fun. I’m essentially training a digital assistant to work the way my brain works, rather than forcing my brain to work the way some app designer thought it should. Over time as this personal knowledge base grows and LLMs become even more capable, this could have a lot of interesting applications.

What’s Next

I’m still iterating on this. Some things I want to figure out:

  • Deeper Notion integration. Can the MCP bridge the gap, or is Notion destined to stay separate?
  • Cross-project insights. When I learn something in one project, how do I make sure that knowledge transfers to other contexts?
  • More proactive suggestions. Instead of just responding to commands, could Claude notice patterns and suggest things? “Hey, you haven’t journaled in three days” or “you’ve been grinding on this project - maybe take a break?”

If you’re curious about the technical details, the repo is at… actually, it’s private. This is my literal life we’re talking about. But if you’re interested in setting up something similar, hit me up. Happy to share what I’ve learned.