What if you could hand an always-on junior developer a dedicated machine, walk away, and have it writing code, running tests, triaging pull requests, and fixing bugs while you sleep, all for the cost of an old MacBook gathering dust in a drawer? That is the exact setup a developer named YK Dojo recently documented in an exhaustive step-by-step guide that has been making the rounds on Hacker News. The approach is disarmingly simple: take a spare Mac, create an isolated account with zero personal data, give Claude Code SSH access, and let it work independently on its own machine. No containers, no VMs, no cloud monthly bills. Just a laptop you already own turned into a 24/7 engineering assistant.
The Case Against Containers for AI Agents
The obvious question is why not just run Claude Code in a Docker container on your main machine. The guide's creator tried that route first; he even built an entire containerized environment for the purpose. But three limitations pushed him toward a physical spare Mac instead. First, a container still lives on your main machine, so network requests, file system access, and process isolation are never truly separate. Second, containers cannot run Mac-native applications. If you want your AI agent to build a Unity game, run Xcode, or control any macOS-only tool, a container simply cannot do it. Third, computer use, where Claude Code sees the screen and clicks buttons via mouse and keyboard, is severely limited on a container. Running on a separate physical machine with its own display, its own network identity, and its own permissions model solves all three problems at once. The trade-off is that you need a second Mac, but for solo founders who upgrade laptops every few years, the old machine sitting in a closet is often exactly what is needed.
Core Setup: From Old Laptop to Always-On AI Worker
The setup process breaks down into ten well-documented steps that take roughly an hour from start to finish. First, wipe the spare machine entirely: erase all content and settings through macOS Recovery or System Settings, then install the latest macOS version. This ensures the agent has no access to any personal data from the machine's previous life. Next, create a fresh local user account without signing into any Apple ID. This is critical because the agent will have admin privileges on this machine, and isolating it from your iCloud, keychain, and personal files is the entire security model. With the account created, enable Remote Login (SSH) from System Settings or the command line, then configure passwordless sudo so the agent can run administrative commands without getting stuck on a password prompt. The guide includes a one-liner that writes a sudoers rule file, but warns that a typo can lock you out of sudo entirely. Always run sudo visudo -cf to validate the syntax. After that, generate an SSH key on your main Mac and copy it to the target machine with ssh-copy-id, which eliminates password prompts for future connections. The final baseline step is keeping the target awake permanently. macOS defaults to sleeping after about ten minutes of inactivity, which would take the agent off the network. Running sudo pmset -c sleep 0 and sudo pmset -c disablesleep 1 stops system sleep entirely while plugged in, and a defaults command disables the screensaver so the screen never locks.
Clipboard Sync and Claude Code Installation
Once the machine is reachable over SSH without a password, the next layer is clipboard synchronization. macOS ships with pbcopy and pbpaste, which read and write the system clipboard. Piped over SSH, these commands let you push code snippets, credentials, and instructions between machines without any third-party service. The guide wraps this into a single clip.sh script that also supports images, which pbcopy does not handle natively. With the clipboard bridge working, installing Claude Code is a single SSH command: curl -fsSL https://claude.ai/install.sh | bash. The guide recommends pinning a specific version for reproducibility. After installation, the target needs its PATH updated in ~/.zshenv so Claude Code is available to non-interactive SSH sessions. This is an easy-to-miss step that will cause confusing failures later if skipped. The guide also offers an optional setup-claude-env.sh script that installs shell aliases, the GitHub CLI, Playwright MCP for browser automation, and yt-dlp for video processing. Every component is toggleable during an interactive run, or you can pass flags like --all for a hands-off installation. Logging into Claude and GitHub on the target machine requires interactive SSH sessions; the guide walks through the device-code authentication flow, and recommends using a separate GitHub account so the agent cannot accidentally push to your personal repositories.
Computer Use, Phone Control, and Remote Access
The most powerful feature of this setup is computer use over SSH: the ability to have Claude Code see the spare Mac's screen, move its mouse, and type on its keyboard, all driven remotely. This does not work out of the box because macOS gates screen capture and input behind Screen Recording and Accessibility permissions that are tied to the GUI login session, not to an SSH process. The workaround is elegant: a LaunchAgent keeps a tmux server alive inside the GUI session on a fixed socket. Every Claude Code session created on that tmux server inherits the GUI session's permissions, so it can reach the display. The setup-computer-use.sh script automates the entire LaunchAgent installation. Once configured, the ic.sh wrapper script on the source Mac provides a clean command-line interface. The command ic spawns a new session, ic -c continues the most recent conversation, ic ls lists all live sessions, and ic attach lets you jump into a running session. The phone control feature is enabled through Claude's built-in remote-control mode. Running ic rc starts a server that lets you start new sessions and drive existing ones from the Claude mobile app. Combined with Tailscale for remote access, this creates a setup where you can hand off tasks to your spare Mac's Claude Code from anywhere in the world: from a coffee shop, an airport lounge, or your phone on the train. Tailscale provides end-to-end encrypted WireGuard tunnels with no open ports, so the machine is reachable but not exposed to the public internet.
What This Means for Solo Founders
This setup is a blueprint for the solo founder advantage in 2026. The core insight is not technical; it is strategic. An always-on AI agent with full computer access on a dedicated machine can work through tasks that would normally require a second engineer: running test suites, triaging bug reports, reviewing pull requests, updating dependencies, writing documentation, and handling routine maintenance. The total cost is zero beyond the hardware you already own and your existing Claude subscription. The security model is sound because the agent operates on a machine that has no personal data, no Apple ID, and no access to your main machine's network. The guide addresses every real-world friction point that would trip up a solo founder trying this for the first time: the macOS permission prompts that can only be approved in person, the SSH clipboard sync that requires a wrapper script, the tmux workaround for computer use permissions, and the Tailscale configuration for remote access. For founders who have been hesitant to give an AI agent broad permissions on their main development machine, this is the safe middle ground that unlocks the full power of autonomous coding agents without the risk.