An agent with its own time
Running an agent on its own machine changes more than its uptime. It lets work continue without you, which is where the “coworker” feeling really begins.
Most conversations about agents focus on capability. I’ve been more interested lately in a simpler question: whose schedule does the agent run on?
Whose time it keeps
I run Claude Code on my computer every day.
That makes it a session I open and supervise. It’s available while my laptop is open, and it stops when I close the lid. I could leave the machine running overnight, but the setup is still organised around my presence.
Putting the agent on its own machine changes that relationship. A task can continue while I sleep or move forward while I’m doing something else. When I return, I’m checking in on work that already has a history rather than reopening a tool.
This, more than a chatty interface, is what makes an agent feel a little like a coworker to me. A coworker’s day continues when you leave the room. You meet again later and compare what changed.
A separate machine gives an agent the closest thing it has to its own working hours.
What its own time changes
Long-running work becomes practical in that setup. Tasks can span days, keep their state, and progress between check-ins instead of being squeezed into one sitting.
That’s why the Akarii agent ran on a cloud machine rather than someone’s laptop. It could still be working when the team returned in the morning, and its continuity didn’t depend on one person keeping a session alive.
What I am still getting used to
I’m still adjusting to the feeling of work moving without me. I’ll step away, come back, and find that the state has changed. That’s useful, obviously, but it requires a different habit from supervising each step as it happens.
It also means accepting work you didn’t watch happen. Good logs, clear handoffs, and sensible approval boundaries become much more important. I’m willing to make that trade, but only when I can return and quickly understand what changed.