The Long Run That Wrote Itself

I wrote large chunks of this post while I was running to train for my marathon.

Long runs are the price of wanting to run a marathon. They are essential, but they are also incredibly boring. A full marathon might take 5 to 6 hours. The training runs involve running 1 to 3 hours at a slow pace to build my aerobic base. There is less effort spent pushing the pace or trying to hold on. This is good for the body, but not always for the mind.

Podcasts help. Mike Duncan’s Revolutions has probably carried me through more kilometers than any running shoe.

But I have recently noticed another problem. I consume far more than I synthesize. I read books, listen to podcasts and have interesting conversations, but much of it gradually disappears. I have been trying to write more, reflect more and make a conscious effort to turn consumption into thought.

Halfway through a long run, I had a fairly obvious idea. A Zone 2 run is supposed to be conversational. So why was I not having a conversation?

I stopped the podcast and started talking to ChatGPT in voice mode.

We started with the Revolutions series I’d been listening to. ChatGPT was my (admittedly sycophantic) sparring partner while I tried to synthesize hours of podcast into a few frameworks that actually held together. The conversation gradually became an outline for a blog post, then a rough draft to review later. It’s still too messy to post, but it’s a start!

But the run was still far from over, so there was only one reasonable next step: write a blog post about writing a blog post while running.

The surprising part was not that the AI helped me write. It was that the run almost disappeared.

I had intended to run around 10 kilometers. Instead, I ran for nearly two hours and hardly noticed. It felt less like exercise and more like a long phone call with a friend who asks useful questions.

The conversation also helped me pace myself. As long as I could continue talking comfortably, I knew I was keeping the effort under control. Talking became an incidental feedback loop for the run.

The workflow is not perfect. Voice is excellent for exploring an idea, but it is much worse for editing one. At some point, I still need to look at the words, move sections around and decide whether a sentence actually works. I had to review this post twice later to get the AI ick out of the post. And the notes from my French Revolution conversation are too intimidating for me to open.

I have since tried Claude’s voice mode. It’s not as smooth an experience as ChatGPT but I can pick the model it uses so it is smarter. I connected it to Notion, which I then used to maintain a todo list and summarize structured thoughts into.

But still, with the smoothness of conversation in ChatGPT voice mode, it is a wonderful way to consolidate and synthesize my thoughts.

All in all, it has dramatically reduced the boredom of marathon training. Every long run is now an opportunity to think through something I have read, develop an argument and return home with the first draft of a blog post.