She does her best thinking on a 40-minute walk along the river before her first call. She comes back with three or four threads in her head that felt sharp at the time. By the time she has made coffee, opened her laptop and answered the first email of the morning, two of them are gone. The third is a fragment she cannot quite reconstitute. The fourth she writes down, badly, and never looks at again.
Many owner-operators have a version of this. The thinking happens. The capture does not.
This post is about a small loop that closes the gap. Walking, talking out loud into a voice tool, and asking AI to listen back. None of the parts are new. The combination, with a model on the receiving end, is finally cheap and simple enough that it is worth treating as a practice rather than a curiosity.
What is the dictation-and-respond loop?
A three-step practice. You walk for 30 to 40 minutes and talk out loud into a voice tool, with no editing. The tool transcribes the audio. You then hand the transcript to an AI model with a short standing prompt asking it to summarise the themes, surface the questions you left open, and flag where you contradicted yourself. The walk does the thinking. The model plays back what you actually said.
The loop sits in the Do quadrant of the EAD-Do framework. AI is not eliminating work, automating it, or taking it from you. You are still the one doing the thinking. The model is a faster, more patient version of the friend who used to walk with you and ask “what did you mean by that?” when you trailed off.
Why does walking and talking actually work?
Because the body and the voice are doing useful cognitive work that desk-and-keyboard work suppresses. Marily Oppezzo and Daniel Schwartz at Stanford ran four experiments in 2014 and found walking lifted divergent creative output by around 60% on average, with around 81% of participants improving. The effect persisted into the seated work that followed the walk. Walking outside scored highest, but indoor treadmill walking against a blank wall still beat sitting.
The historical record points the same way. Aristotle’s school in Athens was named the Peripatetic school after the walkway its members thought along. Nietzsche, Kierkegaard and Thoreau all wrote, separately and without much common ground otherwise, about walking as the place their best thinking happened. None of them were describing exercise. They were describing a specific cognitive state that walking produces and sitting does not.
Talking out loud adds a second layer. Speaking a thought engages language production, motor systems and your own auditory feedback at the same time, which is why explaining a problem to someone often surfaces the answer before they have spoken. The dictation-and-respond loop is letting that happen on a walk, with a model standing in for the listener.
Where do you actually meet this in your week?
Anywhere you have 30 to 40 minutes of moving time and a problem that has gone stuck on the page. The first natural slot is the morning, before the inbox opens. The second is the post-lunch reset, especially on days the morning has been back-to-back calls. The third is the deliberate “I am stuck” walk, taken mid-task when the work is not landing.
The voice tool can be almost anything. A phone voice memo plus a transcription pass works. AudioPen wraps the same idea into a single record button. Whisper sits underneath many of the better-known products. Wispr Flow and Superwhisper are the current options if you want the transcript to land inside the document you are already working in. The choice matters less than the discipline of using the same tool for a month so the friction stops being the topic.
The standing prompt is the part founders commonly skip. A useful version is two lines: “Read this transcript as a sceptical listener. Give me the three or four themes I kept coming back to, the questions I left open, and the places I contradicted myself, in plain English, no preamble.” That prompt, pasted in front of the transcript, gives you a useful response 90% of the time. Tweak it for your own register, then leave it alone.
When should you ask for the loop, and when should you ignore it?
Ask for it when the problem is shaped like thinking, not execution. Strategic calls where the data is in but the decision has not landed. The tone of an awkward Friday conversation. A piece of writing not flowing. A recurring frustration with someone in the team where you cannot tell if the issue is them, you, or the role. In those cases, walking played back through a model beats an hour at the desk.
Ignore it when the work is convergent and answer-shaped. Detailed financial modelling, code debugging, contract review, anything where the cognitive task is narrowing toward a single correct answer rather than opening into possibilities. The Stanford research found walking gave no benefit, and sometimes a small drag, on convergent thinking. Sit, focus, do the work.
Ignore it also when you have not committed to acting on what comes out. The loop without the action turns into a transcripted version of the same thinking that already used to vanish on the walk back. The simplest discipline that fixes this is acting on one insight per walk that week. One insight, one action, before the next walk. Without that, the practice becomes journaling with extra steps.
Related concepts on the same shelf
The closest cousin is using AI as a sparring partner for hard decisions, where the dialogue happens at the desk rather than on a walk and the model pushes back rather than playing the transcript back. The two practices stack, the walk produces the raw thinking, the sparring session pressure-tests the conclusion.
A second neighbour is treating strategy work as a second brain with AI, the same intellectual register applied to a written knowledge base instead of a walking transcript. Founders who run both end up with a private archive of their own thinking that becomes more useful month by month.
The third is the Monday morning what matters this week practice, which is the natural place for the one-insight-per-walk discipline to land. The walk surfaces the thread, Monday morning chooses what to do about it. Read the pillar post for AI on your own desk and the EAD-Do framework explainer for the wider context, the dictation-and-respond loop is one specific Do-quadrant practice inside that frame.
