Open notebook and fountain pen with handwritten symbols and binary digits rising from the pages, marked with the gold P | L monogram

I don’t see AI’s most valuable personal role as a coach, therapist or chief of staff. I believe it is the possibility of continuity across time and decision levels.

I recently published an article discussing how my journalling habits ended up feeding an AI-driven personal health dashboard. While it was just meant to share some personal experience, it ended up raising more interest than I expected, particularly in the journalling part. So I decided to write a bit more about this, and in fact about how my journalling has evolved in our AI-driven world, and how that has fed into a personal dashboard.

Let me start by going back to an aphorism that has marked me from an early age, attributed to Socrates via Plato: “The unexamined life is not worth living”. The need to look inward, into who I am, what motivates me and what I am becoming has fuelled a writing habit in me since my teens. For years, I journalled with paper and fountain pens. This combination remains one of my small pleasures, and whereas midlife crises drive others towards young lovers and fast cars, mine was an excuse to buy expensive fountain pens. But while paper and pen give me immediate tactile pleasure and make me reflect through the process of writing, their limitation is that paper does not ask follow-up questions, reorganise an argument, compare today’s anxiety with a decision made six months ago or point out that a supposedly new problem has appeared in four previous entries. And it is easy to throw away; in fact, many of my journals ended up in the fireplace, another of my guilty pleasures.

This lack of responsiveness from the paper format has slowly moved much of my journalling towards a continuing conversation with artificial intelligence. The change sounds trivial: instead of writing into a notebook, I write/speak into a dialogue or directly to a Markdown file. But in practice, this has altered what the journal can do and how useful journalling has become to me. I can no longer imagine myself journalling just on paper (does anyone want to buy a fountain pen?).

I thought it worthwhile to make this reflection public, as I see AI currently being offered almost every job in a person’s life. It can be a coach, mentor, therapist, tutor, chief of staff, accountability partner, research assistant, copy editor and personal assistant. I do use it in most of those roles and, actually, as an all-purpose minion. The labels are entertaining, and sometimes useful, but I think they obscure a more relevant aspect. The important development is not that an AI can imitate several advisers, but that it can do all those jobs in a sustained conversation that can connect reflection across time and across domains. A journal feeding such roles can, I believe, become a powerful tool “to live a consequentially examined life”.

My long route through productivity

As a postdoctoral scientist in Cambridge, I read widely and generated project ideas faster than I could complete them, and had an impressive array of datasets and scripts always running, advancing multiple projects. An idle cluster was anathema! At the same time, I trained intensively in Capoeira, played in an ad hoc musical “ensemble”, was developing ideas for my first short stories and novels, and wanted an active social and cultural life. Starting always came easily. Finishing, however, did not. This was the setting in which I became obsessed with Getting Things Done.

GTD was a framework that had a major impact on my life then and now. Capturing commitments reduced the burden of trying to remember everything. Defining projects as outcomes made ambiguous intentions more concrete. Its Weekly Review became a durable habit. Perhaps most importantly, “Someday/Maybe” gave legitimate status to not doing something now. At the time, accumulating unfinished ideas were causing me considerable stress.

Later, back in Portugal and managing both a research group and a bioinformatics core, I used OmniFocus to control a growing collection of projects. I learnt to better estimate the time required for tasks, allocated them more realistically and enjoyed crossing them off. For a while it felt like a magic bullet.

Then the bullet ricocheted. I became very efficient at executing an ever-growing list fuelled by my own excessive ambition, now powered by amazing software. The system helped me do more, but it did not help me decide whether something should be on the list or not, and retrospectively, doing more does not necessarily correlate with achieving more.

The 12 Week Year helped me learn how to prioritise and better schedule my work. A project did not have to sit indefinitely in “Someday”. I could decide that it belonged in the third quarter and exclude it from the first. That reduced stress and improved my output. It also worsened a recurring paradox I already alluded to in the previous paragraph: focus increased productivity; productivity increased confidence; confidence increased ambition; and ambition filled the system again.

My morning practice evolved in parallel. The Miracle Morning initially inspired a rather prescriptive routine involving water, exercise, meditation, reading, writing, affirmations and visualisation. Most of that survived only after I stopped following it as doctrine. Affirmations did little for me. Visualisation gradually became something more grounded: review the direction of my life, consider the current quarter, decide what matters today, and record achievement as well as failure. And because visualising the outcomes I was chasing became a regular exercise, the journalling transmuted into something different, and my journals were no longer simply a record of feelings or events: they were becoming the place where different time horizons met.

Physical planners were an important step in my journey, particularly when I moved from diary to goal planners (if you have not done this yet, try it and get ready for an earth-shattering life change!). But at the same time, they made another problem visible. I initially used an A5 Legend Planner for about two years and later moved to the larger Clever Fox Pro. Legend gave personal and professional priorities comparable space on the page. It forced health, relationships, family and work into the same field of view. Clever Fox added explicit one-, five- and ten-year horizons, and had even stronger granularity across different areas of life. Legend broadened what deserved attention; Clever Fox lengthened the period and broadened the domains over which I considered it.

That physical visibility mattered tremendously to me. A screen tends to show whatever I have opened. My planner spread on the desk showed me the whole week, including the parts of life I might prefer to neglect. Categories and interfaces are not neutral. What receives permanent space becomes harder to ignore.

I am no longer using these planners, but my experiences with my (always blue) Legend Planners, and later with my beloved (always golden) Clever Fox Planners, were truly transformational, and I keep recommending them and offering them to people I see being dominated by task lists, by calendars or by one specific area of life to the detriment of others. But “… Painted wings and giant’s rings make way for other toys, One gray night it happened, Jackie Paper came no more…” *, and I inevitably moved digital in my continuous quest for productivity and self-examination.

* From Peter, Paul and Mary’s “Puff, the Magic Dragon”.

From a record to a conversation

While I’ve been using generative AI tools ever since they became available for specific tasks or searches or “discussions”, their lack of memory precluded any real impact (for me). But when general-purpose AI became capable enough to support serious, persistent dialogue, I started using it more and more. I did not approach it with a clean methodology, but I accumulated uses and somehow translated each of the major learnings, methods and habits.

ChatGPT became a strategic sparring partner. I used it to interrogate plans and develop canvases for the major areas of my life. Claude, connected to my Obsidian vault, became closer to a chief of staff. It helped prepare weekly reviews, track progress and relate current activity to annual and quarterly objectives. I started using Obsidian, which still remains the source of truth under my control. Most importantly, and this is what got me “hooked”, the computer spoke back, disagreed, questioned and forced me to justify, after serious tweaking to avoid as much as possible its tendency to aim to please and agree.

The arrangement has become effective but somewhat baroque. It reflects my preferences, technical comfort and appetite for complexity. I would not recommend reproducing it.

Still, something important happened when journalling, planning and strategic discussion began to share the same underlying record. The AI could compare what I said I wanted with what I repeatedly did. It could notice when an urgent professional project was consuming the capacity supposedly reserved for health or family. It could retrieve the reason a commitment had been made, rather than treating the commitment as self-justifying. A weekly plan could become a test of strategy rather than a more attractive task list. And more importantly, long-term goals had gained a seat at the weekly planning sessions.

I already described how this ended up having a major impact on my health, or more precisely, my own attitude towards it. Over roughly a decade, I assembled laboratory results, clinical reports, genomic findings, wearable data, training records and fragments from my journals into a personal precision-health record. AI made it surprisingly easy to build an interface over that information and explore it as a trajectory. The record took ten years, but the interface that made it clear to me took a single prompt.

The useful insight did not come from a clever dashboard alone (but it does help!). It came from placing measurements beside context: an injury, a change in training, a period of poor sleep, an observation recorded in a journal. The AI could help integrate the data because there was a longitudinal substrate to integrate.

Continuous AI conversation created a similar substrate elsewhere in my life. It contained not only decisions and outcomes, but the reasoning, doubts, conflicts and changes of mind around them. This is where I think the personal significance of AI has been underestimated, at least by me. Its value is not limited to producing a good answer now. It can help preserve the path of a question as it evolves over time with you.

The problem with an AI chief of staff

The fashion for giving AI human titles is understandable. Large-scale usage research published by OpenAI found that about half of consumer messages were “asking” for advice or information, while a smaller but meaningful share involved personal reflection, exploration or play. Anthropic has separately reported substantial personal use of Claude. People are not waiting for a formally defined category. They are already using general-purpose systems across work and life.

But “chief of staff” is a dangerously flattering description if it encourages us to confuse fluency with judgement. A human chief of staff understands institutional history, relationships, tacit boundaries and the consequences of a mistake. An AI works from the context made available to it, interpreted through a model that can be persuasive when it is wrong.

The evidence should make us cautious. A 2026 randomised comparison of accredited human coaches with automated AI coaching found substantial effects only for the human coaching condition. Research in Science across 11 leading models found AI responses were considerably more affirming than human responses, including when users described harmful conduct. In experiments, sycophantic responses made people less willing to accept responsibility or repair interpersonal conflict. A separate study of knowledge workers found that greater confidence in AI was associated with less self-reported critical-thinking effort.

These studies do not prove that personal AI use is harmful, or that AI cannot support reflection. They show why a responsive journal should not quietly become an oracle. Availability is not wisdom. Memory is not understanding. Agreement is not coaching.

The design question is therefore not simply which role the AI should play. It is what structure governs the relationship.

The architecture I extracted from the mess

People close to me saw how I used AI to manage a complicated life, particularly as I navigated complex transitions while holding multiple balls in the air, and asked whether I could make the system useful to them. I was reluctant because the implementation was so personal. Once I tried to remove my preferred tools, review rituals and accumulated quirks, however, the underlying logic became clearer.

I gave the resulting architecture a name: Zullya.

Zullya separates four kinds of thinking that AI conversations often collapse.

Identity asks what future a person genuinely wants within an enduring area of life. Strategy asks which choices, exclusions and commitments could create it. Integration forces the separate strategies for work, health, family, relationships, finance and contribution to confront one another. Operating System translates the integrated whole into reviews, habits, information flows and everyday decisions.

The separation is perhaps one of the most critical aspects, both across the four stages (Identity, Strategy, etc.) and across areas of focus. Identity work becomes timid if every aspiration is immediately reduced to next week’s practical constraints, or if we need to consider our aspiration to write more (guilty!) alongside the reality of the morning family routine. Strategy becomes fantasy if it never confronts conflicts across a finite life. Productivity becomes dangerous when it efficiently advances commitments that no longer serve the person making them.

The AI’s job is to preserve continuity among these levels. It can question, compare, synthesise and maintain artefacts. It can bring an earlier principle into a current decision and bring operating experience back upstream when the strategy needs revision. It does not acquire the authority to decide what somebody’s life should become.

This is my design hypothesis, not a validated method. It works for me; that is the only solid claim I can make. The ingredients are not new: journalling, goal setting, future-self work, coaching, periodic review and task management all have long traditions. AI changes the delivery mechanism by making sustained inquiry and synthesis available at much higher frequency. Whether Zullya’s particular sequence improves decision quality or coherence for other people remains to be tested.

That limitation is central to the claim. The breakthrough is not “AI knows you”. Models can forget, distort, flatter and infer too much. The opportunity is to create a governed record in which the person can see, correct and own the important artefacts, and in which the system is explicitly required to distinguish evidence from inference and challenge from decision.

Plenty of systems now promise an AI coach, an AI journal or an AI life operating system. I am not claiming to have invented any of those categories. My interest is narrower: how can a person maintain continuity between identity, strategy, competing life domains and weekly action without handing interpretive authority to the model? Zullya is my attempt to make that question operational: a collection of files that induce a method of working/reasoning, which can be uploaded to one’s favourite AI to start a personal journey.

What the journal is for now

I still use paper. The pleasure of a fountain pen writing in blue or sepia has not been superseded by the probability distribution of the next token. I tell myself that some reflection benefits from slowness, privacy and the absence of a reply, but maybe I am just getting my next “fix of the pen”. But I admit that, while banal in terms of tactile experience, I increasingly spend time writing on my BOOX, where even my handwriting can become data. But a significant part of my journalling happens in the dialogue box of my AI engine of choice or, even worse, as conversations in the car “with the phone”.

I never really thought of journalling as writing for a future self who may or may not reread it; it was really just a format to look inward, and the writing itself was not necessarily relevant. But now, all that writing (or dialoguing) is data. In its conversational form, it can become a living interface between timescales: what happened today, what this quarter requires, what the next decade is for, and which person is expected to live through all three.

That is my claim on AI. Its most important personal contribution may not be better answers, faster administration or even the simulation of expert roles. It may be the ability to make continuity practical.

The condition is that continuity must remain governed by the person. Otherwise the same machine that connects the record can manufacture coherence where none exists, reinforce a convenient story and help us become more efficient versions of whoever we happened to be yesterday.

My interpretation is that, through using AI, my journal started talking back. The valuable part was not that it had a voice. It was that, for the first time, the journal, the strategy and the week could remain in the same conversation.

Sources and note

When ChatGPT reviewed my text for final polish, it was adamant that I needed to include this, which I hope you find interesting, relevant, fascinating even: Zullya is an unvalidated conceptual architecture developed by the author from his own practice. This article describes personal experience and a design hypothesis, not evidence of general effectiveness. Personal, health and organisational data should be handled with appropriate attention to privacy, security, professional boundaries and user control.