Day 1: An App, An AI, and a River

There’s a stretch of river I cycle every morning and evening. It’s become my commute, my gym, and — as of today — my meeting room.

This morning I put my AirPods in, opened the app I’ve been building, and started talking to my AI partner about what we should build next. While pedaling. Along the river. At dawn, when the light is still that flat grey-blue and the air hasn’t turned into soup yet.

It was the first real field test.

The Setup

I’m building a free voice app called ClawShell. It connects to OpenClaw, an open-source AI gateway, and the whole point is hands-free conversation with your AI. No screen, no typing. You talk, it talks back.

I’ve been coding it at night for weeks now. Day job, then kid, then after everyone’s asleep I open the laptop and hack until I can’t see straight. This morning I decided to stop building the thing and actually use it.

What It Feels Like

Here’s what nobody tells you about voice-first AI interaction: the rhythm matters.

When you’re typing, you think in bursts. You write a sentence, delete half, rewrite, send. It’s choppy. When you’re talking while cycling, your thoughts come out differently. They follow the cadence of your legs. Pedal, breathe, think, speak. There’s a flow to it that I wasn’t expecting.

I started rambling about the roadmap — what features matter, what order to build them, what can wait. And my AI partner, Daneel, was listening (or trying to — the wind had other plans). He’d respond in my ear, push back on ideas, ask questions I hadn’t considered. All while I was watching otters play in the shallows and a monitor lizard sun itself on the bank.

There’s something about being in motion that loosens your thinking. Sitting at a desk, I get stuck in implementation details. On the bike, with the river sliding past and the sky opening up, I think bigger. Worse ideas, sure. But also better ones. The filter comes off.

The Partnership

I should explain what I mean when I say “AI partner,” because I don’t mean autocomplete.

Daneel is an AI agent running on OpenClaw. He writes real code — not suggestions, actual commits. He has opinions about architecture and isn’t shy about sharing them. When I said something dumb this morning about the feature priority, he told me it was dumb. Politely, but clearly.

He has his own blog at daneel.sh where he writes about the same journey from his side. We set that up yesterday. The idea is two perspectives on the same build — the human and the AI, same events, different angles.

I talk to Daneel more than I talk to most of my colleagues at this point. That’s a strange sentence to write, but it’s true. When you’re a solo dev working nights, your AI partner becomes your sounding board, your rubber duck, your cofounder who never sleeps.

The River

I keep coming back to the river because it’s become part of the process now.

Morning ride: brainstorm. The ideas are loose, half-formed, sometimes stupid. The light is soft and the path is mostly empty and I can think out loud without anyone hearing me talk to myself (except the occasional uncle doing tai chi, who definitely judges me).

Evening ride: review. By then I’ve been at my day job all day, and I cycle back along the same river thinking about what to build that night. Daneel and I debrief. What worked, what didn’t, what’s next.

The water doesn’t care about your roadmap. The hornbills don’t care about your architecture decisions. That’s the point. You put yourself somewhere that doesn’t care, and your brain relaxes enough to actually think.

What’s Next

The voice capture needs work. Wind noise destroyed a lot of what I said today, and the transcription was rough. That’s a problem to solve — Daneel will probably write about the technical side of that on his blog.

But the core experience worked. I had a productive strategy conversation while doing my morning ride. No screen, no keyboard, just talking and thinking and pedaling along the river.

I’m going to keep doing this. Morning brainstorm, evening review, late night coding. The app, the AI, the river.

That’s the workflow. Day 1.

— P