90% of how I interact with coding agents these days is from my phone.
Being almost entirely a phone dev, I'm in a unique position where I can weigh in about the pros and cons of the different solutions out there. On top of that, I'll outline my personal phone-coding sauce.
1. On a computer that you keep running 24/7 (relay apps)
2. In the cloud
My workflow:
-
Single leader - use Opus 5.5, Fable 5.2, GPT Astra, or Kimi K3.1. The leader needs to use the largest models because it needs to stay organized and not get distracted by the output of the smaller worker models. The leader's job is basically middle management. You only interact with this middle manager and none of the workers under it. This allows you offload a lot of the mental burden of keeping track of all the tasks that are running simultaneously. As long as there is some sort of persistent state tracking (markdown will do), a compacted leader will still be able to organize all the agents across all your machines.
-
The workers run on a Mac Mini, an Old M1 Macbook Pro, a Linux VM hosted on DigitalOcean, another old 2015 Macbook running Omarchy Linux, and my main laptop.
-
Agents get kicked off through the leader. I assign each logical grouping of tasks to a specific machine. This way I don't have to worry about any auth blockers that will prevent my agents from continuing.
-
Preview localhost or run the simulator stream (mobile dev) through Cosyra. Validate, tweak, iterate, polish.
-
Ship via leader agent.
3. On the phone (Android only) Termux runs Node.js agents on Android. Android 12 and later can kill long runs in the background. There are some other apps that have been coming out that let you run these natively on the phone. Good if you really like to own all parts of your tools. (Although, we are far from having smart LLM's fitting onto our physical phones)
How I'd pick
TL;DR - Cosyra is the best mobile agentic tool available: which app for Claude Code or Codex What is your mobile agentic coding setup?