Everyone knows they ought to write things down. Everyone has tried it. And nearly everyone I meet has three half-used note-taking apps sitting on their phone, each abandoned at a slightly different point of enthusiasm. Mine were Notes, Things and a beautiful outliner I paid for twice. Lovely, all of them. Used properly for about eleven days.

The usual explanation is discipline, which is both unkind and wrong. The better explanation sits in the research on executive function: the set of cognitive processes that plan, initiate, sequence and hold in mind whatever you are trying to do. It is not a personality trait. Everyone has it, everyone spends it, and some people have less of it available on a given Tuesday than others. Every task that draws on it pulls from the same pool.

Which means "just write it down" is not free. It is expensive. And it sends the invoice at exactly the moment you can least afford it, which is halfway through something else.

That is the whole argument of this piece, and the rest of it is the mechanism, the citations, and what the mechanism should change about the tools you pick. I have kept my opinions clearly labelled as opinions.

What executive function actually is

The standard model breaks executive function into a small number of separable but correlated components. Miyake and colleagues established this experimentally in a latent variable analysis in 2000, finding three distinguishable factors: updating (working memory), shifting (cognitive flexibility) and inhibition 1. Friedman and Miyake later summarised a decade of follow-up work under the heading of unity and diversity: the components are related, but not identical 2.

In plainer terms:

  • Working memory. Holding information active while you do something with it. Baddeley and Hitch's multicomponent model is the reference point here, updated by Baddeley in 2000 with the episodic buffer 3 and reviewed by Hitch, Allen and Baddeley fifty years on 4.
  • Cognitive flexibility. Switching between tasks or rules without dragging the old one along.
  • Inhibitory control. Stopping an automatic response so a chosen one can happen instead.
  • Planning and initiation. Working out the sequence, and then, the harder part, starting it.

Two numbers matter for our problem. First, capacity is small. Cowan's review argues the real limit of what you can hold in mind is closer to four chunks than the famous seven 5. Four. Not four documents, four chunks.

Second, in ADHD these functions are not incidentally affected, they are the centre of the account. Barkley's 1997 unifying theory places deficient behavioural inhibition upstream of four executive functions, working memory among them 6. Brown's model, developed over decades of clinical work, describes ADHD outright as a set of executive function impairments rather than a behaviour problem 7, set out at length in his 2013 book 8.

I frame this as executive function rather than ADHD on purpose. It is more accurate. The mechanism below applies to a tired parent at 6pm, a founder on their ninth meeting, and someone with a diagnosis, in different amounts.

For the "write it down" problem, the bottleneck is working memory and initiation. Hold that thought. Ideally somewhere other than working memory.

Why note-taking is more executively demanding than it looks

Here is what "just write it down" actually contains:

  1. Notice you have a thought worth capturing.
  2. Decide it is worth capturing. That is an evaluation, and it costs something.
  3. Decide where it goes. Notes? Todoist? Notion? The one you are trying to commit to this month?
  4. Unlock the phone, switch context.
  5. Find and open the right app.
  6. Navigate to the right note, or start a new one, or worry about which.
  7. Type the thought, in a moving lift, one-handed, with autocorrect having views.
  8. Return to what you were doing, and reconstruct where you were.

Eight operations. If any single one fails, the thought is gone, and you will get it back at 11pm on Sunday with no context attached.

Worse, they compete. Step 3 alone is a decision under time pressure, and Sweller's cognitive load work is the canonical statement of the general problem: extraneous load imposed by the format of a task consumes the same limited resource the actual task needs 9. Every second spent on capture is a second not spent on the next thing, and for the whole of that interval your four chunks are being displaced by app-navigation trivia.

My unpopular opinion: step 3 is the real killer, and it is entirely self-inflicted by tool sprawl.

Why voice capture is a real change, not a marketing line

The same job, spoken:

  1. Notice you have a thought worth capturing.
  2. Tap the widget, or say the trigger.
  3. Talk.

Three operations. No destination decision, no typing, no navigation. And the capture happens while you continue doing roughly what you were doing, because speech shares less machinery with walking down a corridor than typing does. Working memory is not held hostage, because the thought is externalised in about four seconds.

Two supports for this being real rather than rhetorical. Ruan and colleagues found speech input on touchscreen phones to be roughly three times faster than keyboard entry for short English messages 10. And the dual-tasking literature is clear that concurrent tasks interfere in proportion to shared demands, with talking while walking measurably cheaper than harder concurrent tasks: Kemper and colleagues on the costs of doing two things at once 11, and more recent work on cognitive-motor interference during walking, where task type drives the size of the cost 12.

It is not magic. It is a genuine reduction in the number of executive operations between having a thought and having it recorded. That is all, and it is quite a lot.

Why voice-to-text alone doesn't finish the job

Now the part the transcription apps would rather not discuss.

If the voice note becomes text you have to review later, you have not removed the executive cost. You have posted it forward to tomorrow-you, with interest. Tomorrow-you must:

  1. Open the app.
  2. Read the transcript, which is longer than you remember and less coherent.
  3. Decide what to do with it.
  4. Actually do the thing.

Four more operations, and steps 3 and 4 are the expensive ones, because they are decision and initiation, the two functions that were already thin. Deferred is not removed. My folder of 340 perfect transcripts was proof of that, and I will not be showing it to anyone.

The apps that do more, that draft the email, create the task, file the note against the right person, remove operations 3 and 4 for tomorrow-you. The decision is reduced to approving or editing something that already exists, which is a vastly cheaper cognitive act than authoring from scratch.

That is the mechanism. Not the transcription. Everyone can transcribe now, and to a very high standard. What matters is what happens after.

What this means for tool choice

Four practical implications, in order of how much they actually decide the outcome.

  1. Capture friction beats features. An app you can start recording in two seconds from a locked phone will outperform a richer app that takes seven. Count the taps. It is the most useful five minutes of evaluation you will do.
  2. What happens after beats transcription accuracy. A 95% accurate transcript sitting in a folder is worth less than an 85% transcript that became a drafted email waiting in your outbox. Accuracy is table stakes; disposition is the product.
  3. Automatic surfacing beats search. If you have to remember to search, you will forget to search, because remembering to look is itself an executive operation. Tools that push things back at you as drafts, tasks or calendar holds do better for exactly this reason.
  4. Widgets and quick actions matter enormously. Every extra tap between the thought and the recording is another place the chain can break. Lock screen, home screen widget, action button, whatever your phone offers. Set it up on day one, not eventually.

What this can't do

Executive function is trainable to a degree and limited in any case. Voice-first tools reduce demand on it for one specific task: capture. They do not:

  • Increase your overall executive function capacity.
  • Fix task initiation on work you are avoiding for reasons that have nothing to do with capture.
  • Substitute for medication, therapy or coaching for ADHD. Nothing on your home screen does that, and anyone selling otherwise should be ignored politely and then firmly.
  • Solve time blindness, motivation, or the fact that you agreed to too many things.

The mechanism is real. The scope is narrow. Voice capture is a lever on one problem, and it is worth being precise about which one, because overclaiming is how this whole category lost its credibility.

The point, briefly

"Just write it down" is bad advice not because people lack discipline, but because writing things down carries a genuine cognitive cost, and on a full day that cost is higher than many people can afford. Eight operations, drawn from a pool of four chunks, at the worst possible moment.

Voice capture does not fix the person. There is nothing to fix. It cuts the cost, from eight operations to three, and if what you say then becomes a draft rather than a transcript, it cuts tomorrow's cost too.

That is the bet behind Agentsy, which is my company and which I build, so weigh the enthusiasm accordingly. It's free while you're testing it, and if you disagree with the mechanism I have described here I would genuinely like to hear why: hello@agentsy.ai.

References

  1. Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., Howerter, A., & Wager, T. D. (2000). The unity and diversity of executive functions and their contributions to complex "frontal lobe" tasks: A latent variable analysis. Cognitive Psychology, 41(1), 49-100.
  2. Friedman, N. P., & Miyake, A. (2012). The nature and organization of individual differences in executive functions: Four general conclusions. Current Directions in Psychological Science, 21(1), 8-14.
  3. Baddeley, A. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Sciences, 4(11), 417-423.
  4. Hitch, G. J., Allen, R. J., & Baddeley, A. D. (2024). The multicomponent model of working memory fifty years on. Quarterly Journal of Experimental Psychology.
  5. Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24(1), 87-114.
  6. Barkley, R. A. (1997). Behavioral inhibition, sustained attention, and executive functions: Constructing a unifying theory of ADHD. Psychological Bulletin, 121(1), 65-94.
  7. Brown, T. E. The Brown model of executive function impairments in ADHD.
  8. Brown, T. E. (2013). A New Understanding of ADHD in Children and Adults: Executive Function Impairments. Routledge.
  9. Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257-285.
  10. Ruan, S., Wobbrock, J. O., Liou, K., Ng, A., & Landay, J. A. (2017). Comparing speech and keyboard text entry for short messages in two languages on touchscreen phones. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 1(4).
  11. Kemper, S., Herman, R. E., & Lian, C. H. T. (2003). The costs of doing two things at once for young and older adults: Talking while walking, finger tapping, and ignoring speech or noise. Psychology and Aging, 18(2), 181-192.
  12. Pitts, J., Singhal, K., Apte, Y., Patel, P., Kannan, L., & Bhatt, T. (2023). The effect of cognitive task, gait speed, and age on cognitive-motor interference during walking. Sensors, 23(17), 7368.