In February 2026, Anthropic shipped remote control for Claude Code. Type /rc in a local terminal, or launch a session with claude --remote-control, and the terminal throws up a QR code. Scan it with the Claude mobile app and you can watch that local session in real time, even step in from your phone to approve or redirect its next move. In August, the feature got a key reliability update: any machine you've signed into now shows up as a device card, and you can start a brand-new session on an idle machine straight from your phone, not just take over one already running. That sounds like a small change, but it means coding agents have, for the first time, fully decoupled a continuously running session from a human glued to a screen.
OpenAI took a different road to a similar place
OpenAI's Codex built a similar experience inside the ChatGPT app, letting users launch and track Codex tasks straight from their phone. GitHub went further, making both Claude and Codex available as coding agents for Copilot Business and Pro customers, running directly inside github.com, GitHub Mobile, and VS Code. But this path has an obvious limit. Official remote control tools are generally tied to their own ecosystem and account system, Claude's solution serves Claude Code, Codex's solution serves Codex, and neither offers a unified, cross-vendor dashboard where you can manage multiple agent sessions running on different models from the same screen.
The Model Context Protocol (a standardized way for AI models to call external tools) has been integrated into ChatGPT, Cursor, Gemini, Copilot, and VS Code, laying the infrastructure groundwork for cross-vendor agent collaboration.
The real liberation isn't typing on a phone. It's not having to watch the screen anymore
Most people seeing Claude Code or Codex on their phone for the first time assume the point is being able to write code on a phone, but that's a misread, typing code on a phone screen is never going to feel better than on a laptop. The actual productivity leap comes from push notifications replacing the act of sitting and watching an agent work. In the old model, you kick off a task and then have to stay at your desk, watching it execute, worrying it'll get stuck, ready to step in at any moment. The new model looks like this: you start a task on your computer, close the lid, go to a meeting or get lunch or do something else entirely, and when the agent hits a point where it needs your judgment, your phone buzzes. You open it, review the diff it generated, tap approve, and it keeps going. After the August update, the flow can even run in reverse: you're out somewhere, remember something you need the office machine to handle, and start a fresh session on that device card straight from your phone, no need to run back and open a terminal first. That shift to asynchronous work is the actual core value of the 2026 mobile agent wave, not the change in screen size.
This pattern isn't just for writing code
Once you break it down this way, it becomes clear the pattern has nothing inherently to do with programming. Any long-running agent task that occasionally needs human judgment fits the same logic. A competitor-pricing monitor agent scrapes data continuously in the background and only pings you when it spots an anomalous swing. A contract-review agent batch-scans forty-seven documents and only needs your signature when it hits a high-risk clause. A social-content agent has already drafted next week's publishing schedule and just wants you to glance at the headlines on your phone and nod, or tweak one word. Cross-device sync, continuous execution, interrupt only when needed, this combination is really what turns an agent from a passive chat window into a digital coworker with its own autonomous rhythm. For an AI avatar platform like KarmaBox, that means your avatar doesn't need you staring at your phone the whole time. It keeps running in your pocket, and only knocks gently when it genuinely needs you.