A travel journal, not a form to fill in

I tell the agent what I did today. Sometimes I add how I feel, what made an impression on me, or a thought that came up during ordinary activities. I do not have a fixed set of questions that I work through point by point every day. It is more like a travel journal — except the journey is my life, not a holiday.

AI's role is to help me record that story consistently. The agent determines the date and organises the content instead of making me open a form and select one field after another. The organisation of the record should be repeatable, not what I have to say on a given day. I want to describe an experience, not administer a notebook first.

Observations matter most to me. They lead to conclusions: sometimes a sudden “aha”, sometimes a thought that matures across several situations. If I write nothing down, it is difficult later to reconstruct all the elements from which that conclusion emerged. A note preserves at least what I noticed and could put into words at that moment.

That does not mean recording every activity. The mere fact that I ate a sandwich is usually of no use to me. If nutrition became an area needing attention, the level of detail could change. At present I mainly record activities, how I feel, plans and reflections. I do not want to turn life into an endless report on operating my body.

One map, different places for information

My notes live in Obsidian. Its vault is a folder containing local text files in Markdown format. This matters: the record can also be opened and edited with another tool, and taking notes itself does not require AI. The agent helps with organisation; it is not a prerequisite for the journal's existence. (Obsidian, “How Obsidian stores data”)

Not everything should remain in a note, however. In my setup, Obsidian holds context and reflections, tasks go to the appropriate lists in ClickUp, and files go to folders in Google Drive. When I say I have something to do, the agent should record the task in the right place without unnecessary questioning. An account of an experience and an obligation to fulfil can emerge from the same conversation, but they serve different purposes.

To keep these places from becoming three separate worlds, I use the same categories throughout: relationships, leisure and experiences, learning and exploration, health and energy, finances, home and everyday life, creativity, work and independence, and “self and direction”. These nine areas give me a map for organising a reflection, a document or a task.

I was inspired by Fryderyk Karzełek's Hexagon of Happiness, which I learned about online. Building on that inspiration and my own needs, I developed my own categories with AI's help. I prefer to call this a “nonagon of harmony”. That name better captures my goal: caring for several interconnected areas rather than chasing an uninterrupted feeling of happiness.

This is my map, not a scientifically established number of parts of human life. I think these areas may resonate with other people too, but that does not oblige anyone to adopt my names or tools. I also leave myself room to change the arrangement. It should help me navigate life, not require me to fit life into folders.

The journal, interview and blog do different jobs

I currently use three scheduled processes in this vault: the journal, the interview and the blog. Each has a different role. The journal process starts the day by checking tasks; recent entries are meant to help the agent choose the tone of the conversation and notice what needs attention. Then I can simply talk about my day.

The interview gathers more lasting context about me. The agent asks simple questions, takes existing answers into account and follows up on different areas of life. The journal mainly says what happened; the interview also helps describe what may not emerge from an account of a single day — how I think, what I need, or the direction I want to take.

The blog is a separate editorial process. I propose topics and answer questions, while the agent organises the material and checks sources. This is not about looking only for confirmation of my observations. If sources challenge my conclusion, I want to know. Editing should also tidy up my sometimes strong language without replacing my thoughts with a text I would struggle to put my name to.

Only a draft I have read and approved can move on to publication on the blog. The private journal is not an automatic reservoir from which everything can be disclosed. It may provide an idea: I notice something repeatedly, explore the topic and want to write about it publicly. But that is a different decision from recording an experience for myself.

This separation also protects against pressure to produce content. Not every thought must immediately become an article, nor every bad day material for an audience. Some entries should remain an ordinary trace of life. Only some may become something worth sharing.

A memory aid, not a machine for inventing the past

External support for thinking did not begin with agents. Psychology has the concept of cognitive offloading: reducing cognitive demand through actions that lower a task's requirements on our cognitive resources. Risko and Gilbert's review discusses the mechanisms and consequences of such behaviour. This provides context for the idea of using written records as support, but it is not evidence that my particular system improves memory or quality of life. (Risko and Gilbert, 2016)

“Operational memory” in the title therefore describes the role notes play in what I do, not a biological function taken over by software. I do not want to remember every detail. I want to be able to return to what I wrote and see what I was doing, planning or thinking at the time.

The agent should work with my context rather than fit me to stories about other people encountered online. General knowledge can help it ask questions and examine conclusions. It is not evidence, however, that I experienced something or felt a particular way. When something is absent from the notes, missing information should not be replaced with a convincing story.

Nor is my own note an infallible record of reality. It preserves my account, including the knowledge and judgement I had at the time. That is why I need to distinguish the original record from later interpretation. If the agent proposes a conclusion, it should be possible to return to the specific entries from which it derived it. Otherwise, an orderly text may merely conceal that we have lost the basis of the reasoning.

What I already do, and what I still want to test

Today I use conversation as a way of recording and organising information. Weekly or monthly summaries, however, are still a planned use — I have not used them yet. I can imagine their value: recalling an important event, bringing together recurring observations, looking at changes in plans. But I do not yet have grounds to describe this as an achieved, long-term result.

I see Aviary, which I am still building, in a similar way. The saved context is meant to help my future assistant talk with me without learning everything from scratch. I am not describing a finished digital identity here. For now I am collecting material that could give it a starting point.

I also take care not to make the record depend on a single laptop. The agent records changes in Git history and sends them to a remote repository. In Git terminology, recording a change locally is a commit, while transmitting commits to a remote repository is done through a push. Saving history locally does not, by itself, mean there is already a copy outside the computer. (Chacon and Straub, “Working with Remotes”)

If the laptop is lost, what matters is a successfully transmitted copy stored elsewhere and the ability to regain access to it. This is no guarantee that data can never be lost. It does provide a concrete reason to check the outcome of saving and sending instead of assuming the agent must have done everything.

A message to my future self

My reason for maintaining this system is simpler than its technical surroundings. I want to organise my thoughts, feel less lost in everyday life and leave behind more than an elusive impression that a lot happened. Perhaps some observations will help someone else. Perhaps one day they will come together as a book. I do not have to settle that now.

I treat writing as communication between the present and the future. Today's note does not need to know tomorrow's answers. It needs to preserve enough of my context that I will not have to guess everything from the beginning later. AI helps me organise that trace; it should not add a journey I never took.

Sources

✦Building systems for greater autonomy.