I'm Terrible at Organizing My Life. So I Built an AI That Isn't.
How a folder of plain-text files — a biography, a CV, a pile of voice notes — quietly became the system that plans my days.
Organization, self-organization, was always my weak side. I was never particularly good at it. And yet I always loved making plans — thinking about what I’d want to do this week, this month, next month, this year. But the planning work itself always came hard to me. I had to set aside a lot of time for it. I used to be amazed at people who can do it on the side, who have everything organized. That’s not me. When I felt that the organization of my life had become indefensible — clearly time to put things in order — that’s when I’d sit down, write some text files, analyze them, write plans, and so on. But that process is resource-heavy, it takes a lot of time, so I did it rarely.
I kept thinking: it’d be great to have some software, some system, that helped me do this — something that took on most of the effort. But I never found such a system, or I couldn’t make myself use it. And here’s what I noticed about any system that tries to help me organize my life: it’s built by someone else, on someone else’s principle, and it often doesn’t fit me. Adapting my own process to someone else’s paradigm feels inefficient to me. Maybe it’s the right path for someone, but it was always hard for me. For example: however many finance apps I tried, I ended up running my finances in Excel — well, in a Google Spreadsheet — because it’s simply the most flexible way to do it. No app gives me the flexibility my process needs. They all do roughly the same thing, but through a paradigm that feels unnatural to me.
It turned out my setup progressed gradually. At first I just wanted a specialized AI that didn’t just answer my questions, but answered them taking into account who I am — all the information about me, what I live by, what I do, what my work consists of. So I made a folder and put a few files in it. One is a biography — I wrote my life story, didn’t generate it, wrote it in English, focused on the moments that I think shaped me, the ones that matter for giving me good advice. I added my CV. What else… that was basically it.
The point was to ask general questions — like, should I take this job or that one, or any choice — and just get answers more specific to me. So I wrote an instruction in CLAUDE.md: before you answer, read who I am, read my biography, read my CV, understand who you’re talking to, and answer from there. And surprisingly, this thing worked pretty well. The answers were personalized right away — I didn’t have to feed in piles of corrections. Often I got the answer I needed from the first prompt.
Then I started recording voice notes and transcribing them into that same folder. I have a system set up that does all this automatically, on an open-source stack, on my own hardware, with no external tools, no cloud services — but that’s another story. The short version: I record a note on my phone and it ends up in the folder I run Claude from for personal questions.
So I started using it that way. One time I decided to ask: “Listen, read my note and tell me what you think. What actions do you think I should take, what should I focus on, what am I maybe doing wrong, what could I improve?” And I’d get all kinds of answers. The feedback made me stop and rethink. What I really liked: the AI basically told me back what I’d already said, but in other words — and it pulled out facts and cross-referenced them, like a really good detective, someone who works with scattered information and has to connect it. It found those facts in my notes.
And the cool part: it’d say, “Look, this week you’ve already said three times that you don’t like this work — maybe, if you dislike it that much, you should work out some criteria to decide whether you want to stay.” And I’m like — damn, really. I’ve been turning this over all week, and the whole time I’m telling myself it’s probably nothing, just a bad mood, or I didn’t sleep enough. When in reality this might be the thing wrecking my life. It’s not that these one-off answers become the basis for my decisions — but it helps me process my own thoughts at scale, because I might not remember what I was thinking or saying a week ago, let alone a month ago, especially under a mood. But my Obsidian vault remembers everything, and Claude analyzes it perfectly.
Eventually I asked it: “Can you build me a plan for the day? You roughly know how things are with me — I told you in the note what I need to do.” Then I thought: “Maybe you can connect to my Google Calendar and look at my email, and update the plan based on that.” All these connectors, MCP — they’ve existed for a while. The question is how to use them usefully for yourself. And the result is, I really like it when someone builds the day’s plan for me. I review it. Early on I’d often say, “No, this is junk, doesn’t work — I need more time on this, less on that.” But the automatic plan-building got good enough that I now largely rely on it. The final decision still stays with me. To me the important thing in planning is that a plan exists — it’s never executed 100%, but the fact that it’s there delivers the main value.
Then I’d say, “Can you write me a plan for the week?” — because that’s always been hard for me. I can manage figuring out a day; a week is harder; a month harder still. But you can ask the AI to write a month’s plan, even a year’s — especially once a critical mass of my notes builds up. By then it basically knows what might be useful to me. By the way, besides the biography, one of the main files I wrote is my long-term plans — what I’d want over a five-to-ten-year horizon. I noted that this is how I see it now, it’ll definitely change over time, but it’s worth treating as a snapshot and looking at it through the prism of evolution.
All of this grew into a skill I called “Morning Processing.” Every morning I run Morning Processing, and Claude reads all my notes from the past day, checks all my calendars, builds the day’s plan, updates the previous day’s plan — looks at what I did and didn’t do, decides what to push to tomorrow, what to drop entirely, what to move not to tomorrow but the day after. It finalizes the old plans and creates new ones. On top of that, it reads my fitness data, my training data — if I slept little, it factors that into the plan; if I slept well, it factors that in too. It knows what training I did yesterday. It even plans my training — but that’s another story.