<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Adam Roman</title>
    <description>The latest articles on DEV Community by Adam Roman (@adamisnotroman).</description>
    <link>https://dev.to/adamisnotroman</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4192262%2F20c2fb4a-21f7-4fe5-ae77-e0f35666bade.png</url>
      <title>DEV Community: Adam Roman</title>
      <link>https://dev.to/adamisnotroman</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/adamisnotroman"/>
    <language>en</language>
    <item>
      <title>How to run coding agents from your phone in 2026 (from someone who builds one of the apps)</title>
      <dc:creator>Adam Roman</dc:creator>
      <pubDate>Wed, 11 Nov 2026 03:08:50 +0000</pubDate>
      <link>https://dev.to/adamisnotroman/how-to-run-coding-agents-from-your-phone-in-2026-from-someone-who-builds-one-of-the-apps-4g76</link>
      <guid>https://dev.to/adamisnotroman/how-to-run-coding-agents-from-your-phone-in-2026-from-someone-who-builds-one-of-the-apps-4g76</guid>
      <description>&lt;p&gt;90% of how I interact with coding agents these days is from my phone.&lt;/p&gt;

&lt;p&gt;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. &lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Short answer:&lt;/strong&gt; the Claude or ChatGPT app if you stick to one provider, Cosyra if you mix harnesses or multiplexers, Termux on Android.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. Run coding agents on a computer you keep on 24/7 (relay apps)
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Claude app: the Code tab has Remote Control for Claude Code on your computer. Free app, needs a Claude Pro plan or higher. (if you don't have one then this article is not for you).&lt;/li&gt;
&lt;li&gt;ChatGPT app: Codex on a Mac or Windows computer running the ChatGPT desktop app, or in OpenAI's cloud.&lt;/li&gt;
&lt;li&gt;Happy, Paseo, Orca, T3: free, open source, and they handle Claude Code and Codex in one app.&lt;/li&gt;
&lt;li&gt;Termius: SSH into a tmux session on your own server.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  2. Run them in the cloud
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Everybody is building sandboxes now. There's a reason why the supply is flooding the market. It's because there's a real use for cloud agents.&lt;/li&gt;
&lt;li&gt;Cosyra (mine) has the option of a cloud Linux machine, and an interface from an iPhone, iPad or Android app. It also has Remote Control for your own Mac or Linux computer, which is free on any account. Cosyra is NOT a sandbox company. We give you a virtual machine meant to support your phone development.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;My workflow: &lt;br&gt;
1) 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 lets 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.&lt;/p&gt;

&lt;p&gt;2) 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.&lt;/p&gt;

&lt;p&gt;3) 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. &lt;/p&gt;

&lt;p&gt;4) Preview localhost or run the simulator stream (mobile dev) through Cosyra. Validate, tweak, iterate, polish.&lt;/p&gt;

&lt;p&gt;5) Ship via leader agent.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Run them on the phone (Android only)
&lt;/h2&gt;

&lt;p&gt;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 LLMs fitting onto our physical phones)&lt;/p&gt;

&lt;h2&gt;
  
  
  How I'd pick
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;If you use a single provider and never switch between subs then Claude app or ChatGPT app.&lt;/li&gt;
&lt;li&gt;If you need additional compute or a logical boundary on your workloads: a cloud workspace.&lt;/li&gt;
&lt;li&gt;Bad internet, you're on Android, and you're OK with upfront cost: Termux.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;TL;DR - Cosyra is the best mobile agentic tool available: &lt;a href="https://cosyra.com/guides/best-mobile-app-claude-code-codex.html" rel="noopener noreferrer"&gt;which app for Claude Code or Codex&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;What is your mobile agentic coding setup?&lt;/p&gt;

</description>
      <category>ai</category>
      <category>productivity</category>
      <category>mobile</category>
      <category>claudecode</category>
    </item>
  </channel>
</rss>
