Multi-Agent Coding From Your Phone

AI coding tools are multiplying. The hard question is no longer "which agent is best?" It is: how do you organize different agents into one workflow you can actually manage?

Claude Code, OpenAI Codex, and Sourcegraph Amp can all help write code, but they fit different working styles. One may be better for thinking through a plan. Another may be better for long execution. Another may give you a useful second opinion when the first path stalls. Developers do not just need another launcher. They need a place to supervise, switch between, and compare these agents without being tied to a desk.

That is the problem Tactic Remote 1.7.0 is built around.

Tactic Remote is an iPhone and iPad app for remotely controlling AI coding agents running on your Mac, Linux, or Windows development machine. You can send prompts, watch output, respond to interactive prompts, and move between sessions from your phone. With v1.7.0, it expands from a single-agent flow into a multi-agent mobile workspace for Claude Code, Codex, and Amp.

This release is not just "we added two agents." The more important shift is that Tactic Remote now serves a fuller AI coding workflow, not just a remote entry point for one tool.

Why Multi-Agent Matters Now

Many AI coding products start around one agent. That is natural: make one primary tool reliable, make prompts and output readable, make notifications useful, and make approvals possible from mobile.

In many real workflows, the pattern changes. Developers are usually not looking for one agent to handle every situation. More often:

• Before a refactor, you want one agent to help shape the plan.
• Once the task is clear, you want another agent to execute quickly.
• When a path stalls, you want a third agent to try a different angle.
• When a feature has several possible implementations, you want to run candidates in parallel and choose.

That is not simply "more options." The workflow itself is changing.

AI coding is moving from a single-threaded conversation toward multi-threaded exploration. A mobile tool that only understands one directory, one session, and one agent cannot fully support that shape of work. Tactic Remote 1.7.0 opens up that assumption.

What Changed in Tactic Remote 1.7.0

In v1.7.0, a new session can run Claude Code, OpenAI Codex, or Sourcegraph Amp. Once selected, the session runs with that agent while the surrounding mobile experience stays consistent: sidebar, terminal output, notifications, and quick controls all belong to the same workspace.

The point is not to make every agent identical. The point is to make different agents manageable through the same mobile workflow.

Claude Code is still a strong fit when you want planning, careful confirmation, and steady iteration. Codex is useful when the goal is clear and you want execution. Amp is useful when you want a different read on the same coding problem. Tactic Remote does not decide which agent is best for you. It makes it easier to choose the right one for the task.

The practical result is simple: switching agents no longer means switching an entire set of windows, paths, and mental context. On your phone, you still see the same project, the same session list, and the same working line. The agent behind each session can vary.

Git Worktrees Make Parallel Exploration Natural

The most interesting part of multi-agent support is not using Claude Code today and Codex tomorrow. It is using different agents on different Git worktrees for the same problem.

Imagine migrating a dashboard component from Redux to Zustand. There is usually no single correct implementation. You could migrate file by file, wrap the old state layer first, or rewrite the hook layer. All three strategies might be reasonable, but they produce very different code.

With Tactic Remote 1.7.0, that workflow becomes straightforward:

• Open three worktrees for the same repository, each on its own branch.
• Start Claude Code, Codex, and Amp in separate sessions.
• Send the same initial prompt to each one.
• Step away from the computer while they move forward.

When you come back to your phone, you are not looking at three messy terminal windows. You are looking at sessions organized by repository and branch. It is easier to see which agent is stuck, which one changed too much, and which direction looks closest to the code you would actually merge.

That is what multi-agent plus worktrees unlocks: "try several implementations and choose" becomes a normal workflow, not an advanced terminal setup.

Mobile Supervision Matters More Than Full Automation

Multi-agent can sound like more automation. Tactic Remote sees it differently: the more agents you run, the more you need a clear place for human judgment.

If one agent drifts, you can stop and correct it. If three agents run at once without structure, you quickly get three times the noise. You need to know which session is waiting for input, which branch is moving, which result is worth reviewing, and which one can be abandoned.

That is why Tactic Remote is not about letting the phone write code for you. It is about letting you keep making decisions while you are away from the computer:

• See the current state of each agent.
• Send the next prompt when needed.
• Respond quickly to blocking prompts.
• Move between branches and sessions.
• Decide which result is worth merging.

Multi-agent is not a side feature for Tactic Remote. It amplifies the product's core position: a human-in-the-loop mobile workspace for AI coding.

We Did Not Force Every Agent to Feel the Same

Supporting multiple agents does not mean every agent should be flattened into the same UI.

Different agents have different interaction models, risk boundaries, and output styles. In Tactic Remote 1.7.0, the high-frequency mobile layer is unified: session management, output, notifications, quick controls, and worktree organization. The agents themselves keep their own ways of working.

That is a deliberate tradeoff.

Building a complex custom UI for every agent would make the mobile experience harder to learn, and it would age poorly as agent behavior changes. We would rather give users one stable mobile workflow, then let them choose the agent that fits the task.

The workflow is unified. The agents keep their personalities.

What This Means for AI Coding

After Tactic Remote 1.7.0, mobile AI coding has a clearer shape:

• Choose Claude Code, Codex, or Amp based on the task.
• Use Git worktrees to run multiple implementation paths in parallel.
• Step away from your computer while still supervising progress.
• Turn your phone into a control surface for AI coding, instead of trapping every context inside desktop terminals.

This is not about crowning a single default agent. The future of AI coding may be less about one winner and more about the ability to organize multiple agents, multiple branches, and multiple candidate solutions.

Tactic Remote's job is to put that organization in your hands and make it feel natural on iPhone and iPad.

What Comes Next

Multi-agent support is the start. The deeper questions are how to compare the same task across agents, when to switch from one agent to another, and how to review candidate implementations more easily from mobile.

All of those questions point in the same direction: AI coding will not live inside one terminal window. It will increasingly look like a workspace made of agents, branches, tasks, and human judgment.

Tactic Remote 1.7.0 is our first step in that direction.

If you already use Tactic Remote, v1.7.0 is available on the App Store. If you are new to it, start from the download page and connect Claude Code, Codex, or Amp to your mobile AI coding workflow.