You Have Notes. You Do Not Have a Decision.

Something in the meeting is now writing everything down, perfectly, and it is not you. What that costs is small, and it compounds, and it lands on the only decisions that matter.

There is a folder on your machine holding every meeting you ran this year, and you did not write a word of it. The bot joined before you did. It transcribed the whole thing, labelled who spoke, pulled out the action items, and had a summary in your inbox before you reached the kitchen. Nothing was missed. Nothing was lost. And when somebody asks you three weeks later what you actually decided, you open the file and read it like a stranger reading somebody else's mail.

You were not slacking. That is the part worth sitting with. You were present, you were listening, you were looking at the person instead of hunching over a keyboard, which is exactly what the tool was sold to you for. And it delivered. The problem it solved was real.

There is a ladder here, and most of us have climbed the whole of it in about fifteen years. A pen catches maybe a third of what is said in a room, so you are forced, continuously, in front of the person you are listening to, to choose which third. Then the keyboard arrived, fast enough to keep pace with speech, and the choosing quietly stopped being necessary. I remember typing furiously at Dow Jones during meetings, capturing everything said as if the words were money itself. Now the notetaker has arrived, and the attending has stopped being necessary too.

The mechanism is not distraction. It is more specific than that, and worse. Knowing a thing is being captured is itself what stops you taking it in. Linda Henkel ran the cleanest demonstration of this at Fairfield. She walked people around an art museum and had them photograph some objects and simply look at others. Afterwards they remembered the photographed ones worse. Fewer objects, fewer details, and less about where in the building they had been standing when they saw them.

The part of that study nobody quotes is the part that matters here. When she asked people to zoom in and photograph one particular detail of an object, the impairment vanished. Not only for the detail they had framed, but for the whole object, including the parts they had deliberately left out of the shot. Choosing what to capture was itself the act of paying attention. The camera was never the problem. Handing it the decision was.

Which is roughly what a team at Michigan found when they tested AI note-taking head on. Three settings: a raw transcript, a live summary, and full automation where the notes are written for you. The people given full automation scored worst on the material they had just sat through. The people given the middle setting scored best. Then they were asked which they preferred, and they chose the one that had taught them least, because it felt easiest. Nobody has yet run the study that matters most, which is what a notetaker does to the people sitting in the room rather than watching a lecture.

Comprehension was never quite what you were doing in that room anyway. You were not revising for an exam. You were deciding.

I cannot decide anything important at a computer. I have known that about myself for years and I have stopped fighting it. When something matters I get up and go somewhere else, and I take a pen, or the digital pen of a Kindle Scribe, and I draw the options out until they have a shape I can look at. Many people working with me have deciphered child-like drawings and my illegible writing. Expected value on one side. What happens if I do nothing, which is the option most people never write down at all. Then I invert the whole thing and argue the opposite case to find what I have missed. If it is genuinely large I add the other ingredients. I run with it, literally, for hours. I leave it overnight to stew. I take it to people I trust, and I go back to the research to see what else is there. Somewhere in the middle of that I ask whether I am even answering the right question, which is the check that has saved me most often.

That is how the decision to build RootAstro.org got made. Not to find a retreat centre already standing, not to rent one for a season and see how it went, but to build a physical business from nothing after thirty years of knowledge work done almost entirely in front of a screen, no bodily movement required. It got decided on long runs, walking in the woods, and on an archery range between arrows, with the options drawn out in my own hand and a small number of trusted people telling me where I was wrong. The journey so far has already been worth more than the easier versions I could have chosen.

The computer, for me, is for acting on decisions. It has never once been where I made one.

You do not write things down to remember them. You write them down to find out what you think.

A notetaker is better than you at the first one. It has nothing whatsoever to do with the second.

Which is why the pen survives the arrival of better tools, and why it does not matter whether the pen is plastic and talks to a cloud. The Norwegian study everyone now cites for this, Audrey van der Meer's team at Trondheim, is almost always reported as paper beating computers. It was not that. Both of their conditions were digital. Their subjects wrote in cursive with a stylus on a touchscreen, and that is exactly where the brain connectivity turned up.

The cleanest account of what is actually doing the work comes from Karin James at Indiana, who taught pre-literate children letters three ways: forming them freely, tracing them, or typing them. Only the children who formed the letters themselves developed the adult reading circuitry. Tracing produced nothing, with the same pen, the same paper, the same hand moving through the same shapes. The active ingredient was not the ink or the page. It was generating the form yourself. A stylus keeps that. A keyboard removes it. A bot removes the question.

So keep the transcript. Let the machine do the part it is genuinely better at. But there is a catch the research is honest about: handwritten notes you never open again lose to a summary you actually read. The pen only wins if you go back to the page. Writing it and never returning is the worst of both, and it is what most of us do.

I want to be careful not to dress any of this up as a productivity argument. You will find people claiming that going back to pen and paper lifts performance by thirty per cent, or cuts an hour off your meetings. Those numbers are invented. The honest position I have read is that most of the measured effects here are small. The best review of handwritten notes puts the advantage at about a quarter of a standard deviation, and only when the notes get read again. The best evidence for writing a plan as a specific trigger rather than a deadline, once you correct it for the studies that never got published, is smaller still.

Which is exactly why it is worth doing. Alpha is never a large effect in one go. It is a small persistent edge with an asymmetric cost, applied where the stakes are highest, compounded over time. One of our founders at Blenheim Chalcot bangs the drum on this all the time. The cost here is a pen and the choice not to do average work. Set that against the decisions you would apply it to, which are the ones that determine whether the next two years of your life were worth anything, and the trade is not close. You are not buying a thirty per cent lift. You are buying a slightly better decision, slightly better held, slightly more likely to actually get done, every single time it matters. This is how great performance manifests itself.

There is one exception to the smallness, and it is the thing my own hand does without being asked. When people draw something rather than write it out, they remember it at better than twice the rate. Not a quarter of a standard deviation. Double. Real problems are not lines of text. Organisations, systems, the shape of a decision and its second-order consequences: these sit in your head as things next to other things, with lines between them. A page lets you put them there. A screen makes you type them in sequence, one after another, as though they happened in order.

If you want alpha in how you perform, that is where it is, and it is not in the tooling. Pick up a pen and put the decision in front of your own eyes, in your own hand, before you touch a keyboard.

On the last morning at Root Astro, before anyone drives home, they will write a single page planning the next ninety days in their own hand. One sentence for each of the three bearings. What done looks like at day ninety. The first action, and the moment it starts. Nobody types it. Nothing records it. Nobody leaves holding a plan that something else prepared for them, because a plan you have been handed is a plan you have only read, and the gap between reading a sentence and generating one is among the most reliable findings in the whole of memory research.

That page is the first day of ninety, not the last of three. Everyone leaves with a structured journal built to carry it: the same plan laid out to be worked week by week, written in by hand or imported straight onto a reMarkable or a Kindle Scribe, whichever is the one you will actually open on a wet Tuesday in November. The object matters less than the fact of it. What we are trying to prevent is the thing that happens to every retreat and every offsite ever held, which is that the decisions get made somewhere beautiful and then quietly fail to survive contact with the calendar. It starts with one page on the morning you leave.

A decision you did not write in your own hand is still only an opinion.

Appendix: the evidence, and what it will not carry

Every claim in the essay traces to a source below.

Note: AI research agents led the research that follows and shortlisted the studies for reading by the author. Their selections were red-teamed for inaccuracies and comprehensiveness. Please also do your own research.

Van der Weel, F. R., & Van der Meer, A. L. H. (2024). Handwriting but not typewriting leads to widespread brain connectivity: a high-density EEG study with implications for the classroom. Frontiers in Psychology, 14, 1219945. doi.org/10.3389/fpsyg.2023.1219945

Thirty-six university students analysed, 256-channel EEG. Handwriting produced widespread theta (3.5 to 7.5 Hz) and alpha (8 to 12.5 Hz) coherence across parietal and central regions; typing did not. The handwriting condition was cursive written with a digital stylus on a touchscreen, and the words came from the game Pictionary. There was no paper condition in this study at all, which is why it cannot be cited for paper and can be cited for the stylus. Note the date carefully: the DOI carries 2023 because that is the volume year, but publication was 26 January 2024.

Pinet, S., & Longcamp, M. (2025). Commentary: Handwriting but not typewriting leads to widespread brain connectivity. Frontiers in Psychology, 15, 1517235. doi.org/10.3389/fpsyg.2024.1517235

James, K. H., & Engelhardt, L. (2012). The effects of handwriting experience on functional brain development in pre-literate children. Trends in Neuroscience and Education, 1(1), 32 to 42. Full text (Indiana University)

Five-year-olds learned letters by printing them freely, tracing them, or typing them. Only free printing recruited the adult reading circuit. Tracing did nothing, despite identical pen, paper, hand and movement. Holding everything constant except self-generated production, and watching the effect disappear, is the strongest available evidence that the active ingredient is forming the shape yourself. That is what a stylus preserves and a keyboard removes.

Umejima, K., Ibaraki, T., Yamazaki, T., & Sakai, K. L. (2021). Paper notebooks vs. mobile devices: brain activation differences during memory retrieval. Frontiers in Behavioral Neuroscience, 15, 634158. doi.org/10.3389/fnbeh.2021.634158

Forty-eight participants encoded calendar appointments on paper, on a stylus tablet, or on a phone, then were scanned an hour later. The paper group showed higher hippocampal and visual-cortex activation. Overall memory performance did not differ significantly between groups, and the widely repeated line that paper users "memorised twenty-five per cent faster" is a distortion. The twenty-five per cent comes from the university press release and refers to how long it took to write the calendar entry, not to memory at all. Two authors were employed by the company that funded the study.

Mayer, C., Wallner, S., Budde-Spengler, N., Braunert, S., Arndt, P. A., & Kiefer, M. (2020). Literacy training of kindergarten children with pencil, keyboard or tablet stylus. Frontiers in Psychology, 10, 3054. doi.org/10.3389/fpsyg.2019.03054

One hundred and forty-seven children, seven weeks. The stylus group landed between pencil and keyboard and differed significantly from neither.

Osugi, K., Ihara, A. S., et al. (2019). Differences in brain activity after learning with the use of a digital pen vs. an ink pen: an electroencephalography study. Frontiers in Human Neuroscience, 13, 275. doi.org/10.3389/fnhum.2019.00275

The digital pen produced a greater neural learning signature than the ink pen, but only among participants already familiar with digital pens. No effect in the unfamiliar group. Small study, and it points at something useful: the stylus penalty in other work may be unfamiliarity rather than the medium. A founder who has written on the same device daily for two years is the familiar group.

Flanigan, A. E., Wheeler, J., Colliot, T., Lu, J., & Kiewra, K. A. (2024). Typed versus handwritten lecture notes and college student achievement: a meta-analysis. Educational Psychology Review, 36(3), 78. doi.org/10.1007/s10648-024-09914-w

Twenty-four studies. The source of the essay's "quarter of a standard deviation" figure: taking and reviewing handwritten notes was associated with higher achievement, g = 0.248, while typing wins decisively on volume of notes captured. A second review of the same literature (Voyer, Ronis & Byers, 2022, Contemporary Educational Psychology, 68) finds essentially no difference once device distraction is accounted for, which is why the essay calls the question unsettled.

Henkel, L. A. (2014). Point-and-shoot memories: the influence of taking photos on memory for a museum tour. Psychological Science, 25(2), 396 to 402. doi.org/10.1177/0956797613504438

Museum visitors photographed some objects and simply observed others. They later remembered the photographed objects worse: fewer objects, fewer details, fewer locations. The nuance is the whole point. When participants zoomed in to photograph one specific part of an object, the impairment disappeared, and memory for the parts they had not zoomed in on was just as strong. Selecting what to capture was itself the act of attending. Indiscriminate capture harms; deliberate capture does not. An AI notetaker is maximally indiscriminate. A person deciding what to write down is the zoom lens.

Chen, X., Ruan, K., Ju, K. P., Yap, N., & Wang, X. (2025). More AI assistance reduces cognitive engagement: examining the AI assistance dilemma in AI-supported note-taking. Proceedings of the ACM on Human-Computer Interaction (CSCW). doi.org/10.1145/3757632

Peer-reviewed, and about note-taking specifically. Thirty participants took notes under three levels of AI help: raw transcript, live summary, or fully automated notes. Full automation produced the worst comprehension scores. The middle setting produced the best. Participants preferred the fully automated condition, because it felt easiest. People choose the setting that teaches them least.

Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676 to 688. doi.org/10.1016/j.tics.2016.07.002

Offloading is not good or bad. It is a decision, and the decision is governed by your judgement about your own mental abilities, which is frequently wrong. That is the precise problem with a notetaker in the room: not that it stores things, but that its presence silently reprices your judgement about whether you need to attend. Pairs directly with the Michigan finding above, where people's preference ran opposite to their performance.

Kiewra, K. A., DuBois, N. F., Christian, D., McShane, A., Meyerhoffer, M., & Roskelley, D. (1991). Note-taking functions and techniques. Journal of Educational Psychology, 83(2), 240 to 245. doi.org/10.1037/0022-0663.83.2.240

Unusually well controlled, and double-edged. Taking your own notes and reviewing them beat everything. But reviewing somebody else's borrowed notes beat taking your own notes and never looking at them again. Since most executives never reopen their notes, the realistic comparison is not pen versus bot but "notes I never re-read" versus "a summary I actually read", and the literature does not obviously favour the pen there. This is why the essay concedes the point rather than hiding it: handwriting wins on encoding, and that win is only banked if you go back to the page.

Kosmyna, N., et al. (2025). Your brain on ChatGPT: accumulation of cognitive debt when using an AI assistant for essay writing task. arXiv preprint 2506.08872. arxiv.org/abs/2506.08872

The famous one. It is a preprint, not peer reviewed, with 54 participants, and it has drawn a published methodological critique (Stanković et al., arXiv 2601.00856) arguing it is substantially underpowered for the number of comparisons run. The authors themselves asked journalists not to use words like "brain rot" or "harm" about it. The one result robust enough to quote is behavioural rather than neural: in the first session, 83% of the ChatGPT group could not produce a correct quotation from the essay they had written minutes earlier, against 11% of those who wrote unaided. Do not repeat the EEG numbers.

Zindulka, T., Goller, S., Fernandes, D., Welsch, R., & Buschek, D. (2026). The AI memory gap: users misremember what they created with AI or without. Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems. doi.org/10.1145/3772318.3791494

One hundred and eighty-four participants generated ideas with and without an AI assistant, then a week later tried to identify which were which. AI involvement impaired source memory, most of all in mixed human and AI workflows. For a leader, forgetting whether the judgement was yours is arguably a worse problem than forgetting the content.

Wammes, J. D., Meade, M. E., & Fernandes, M. A. (2016). The drawing effect: evidence for reliable and robust memory benefits in free recall. Quarterly Journal of Experimental Psychology, 69(9), 1752 to 1776. doi.org/10.1080/17470218.2015.1094494

Seven experiments, and the source of the essay's "better than twice the rate" claim: drawing a word beat writing it out in free recall, often by more than double, and the effect survived controls for elaborative encoding, visual imagery and picture superiority.

Why the ninety-day page is handwritten, and blank

Gollwitzer, P. M., & Sheeran, P. (2006). Implementation intentions and goal achievement: a meta-analysis of effects and processes. Advances in Experimental Social Psychology, 38, 69 to 119. doi.org/10.1016/S0065-2601(06)38002-1

Ninety-four tests. Specifying when, where and how you will act, in an if-then form, produced a medium-to-large effect on goal attainment, d = 0.65. The mechanism is that the cue becomes easier to notice and the response fires without needing a fresh decision, which is precisely where intentions normally die.

Sheeran, P., Listrom, O., & Gollwitzer, P. M. (2025). The when and how of planning: meta-analysis of the scope and components of implementation intentions in 642 tests. European Review of Social Psychology, 36(1), 162 to 194. doi.org/10.1080/10463283.2024.2334563

Effects were larger with a genuine if-then format, with highly motivated participants, and when the plan was rehearsed.

Locke, E. A., & Latham, G. P. (2002). Building a practically useful theory of goal setting and task motivation: a 35-year odyssey. American Psychologist, 57(9), 705 to 717. doi.org/10.1037/0003-066X.57.9.705

Roughly four hundred studies, forty thousand participants. Specific and difficult goals reliably beat vague ones, easy ones, and none.

Slamecka, N. J., & Graf, P. (1978). The generation effect: delineation of a phenomenon. Journal of Experimental Psychology: Human Learning and Memory, 4(6), 592 to 604. With the meta-analysis: Bertsch, S., Pesta, B. J., Wiscott, R., & McDaniel, M. A. (2007), Memory & Cognition, 35(2), 201 to 210. doi.org/10.3758/BF03193441