I've always felt that you need to set aside time to think.
No matter how smart AI gets, someone still has to tell it what's worth thinking about. The truly good ideas still have to come from your own mind.
But when you're sitting at your computer, information keeps flooding in. It's rare to have truly quiet time. I've been looking for a way to let those scattered thoughts rise to the surface on their own.
Turns out, running does exactly that.
It's not that running has some magical power. It's just that during work, information keeps pouring into your brain, and your subconscious is constantly organizing it. When you run, you can't do deep work, but the things your subconscious has been organizing bubble up and become new ideas.
Back when I encountered big projects I couldn't figure out, I would go for a run. As I ran, the logical connections between things would sort themselves out.
I've had this habit on and off for years. Lately I've picked it up again, because I have way too many ideas.
Before, when I ran and had an inspiration, I could only try to remember it mentally. By the time I finished, half of it was gone.
Then I found a surprisingly simple solution: turn on voice input in Feishu.
Here's how it works. Use Doubao Input Method (yes, Doubao's speech recognition is surprisingly accurate). In Feishu, tap the microphone icon and you can talk without holding the button. Just speak whatever comes to mind.
The voice is transcribed into text and sent to a Feishu group chat. My random-thoughts-recorder Agent waits there, automatically organizes the content, and saves it.
The bottleneck used to be the "hand" step. Your brain is fast, your mouth slows things down a bit, and typing with your hands slows them down even more. Now I skip the hands entirely — it goes straight from mouth to Agent. From the moment an idea pops up to when it's recorded: a matter of seconds.
And the accuracy of speech-to-text is good enough. Doubao Input Method handles most corrections well, and on top of that, the Agent understands context. Even if you slur a few words, it can usually figure out what you meant.
These off-the-cuff ideas, collected by the random-thoughts-recorder, go down two paths.
One path leads to my blog-writing Agent. It scans the random thoughts for good ideas, turns them into article topics, and writes them out.
The other path leads to my "Lobster Partner" Agent (an Agent that helps me make product decisions). It organizes the information into suggestions like "here's a direction you could explore next." If I accept, it goes into my to-do list. If not, that's fine too.
No inspiration goes to waste. Every day's thoughts are captured.
What strikes me most is how the gap between "having an idea" and "having that idea executed" has been compressed to nearly zero.
Before, if I had an idea, I had to do it myself. Want to write an article? Open the editor yourself. Want to build a tool? Write the code yourself. The more ideas you had, the more they piled up, until eventually you just stopped having them.
Not anymore. Just say the words, and the Agent gets to work. The cost is unbelievably low.
My current rhythm is: while running, I say whatever comes to mind, and it's recorded by voice. My Agent organizes and categorizes it. The writing Agent picks topics from it and writes articles. The partner Agent distills tasks from it.
What I do is simple: spew out ideas while running, then review the results when I get back. Accept or reject — either way, I've given every idea a proper send-off.