There's an assumption I keep seeing threaded into a lot of agentic workflow advice: context switching is no longer a problem.
Hey, you've got agents, now. Just spin one up and get the work done in the background.
I'm not talking about multitasking. I'm talking about breaking your focus on your current task to pursue something else.
This frustrates me because context switching is not a technical problem, it is a human problem. "Spin up an agent" is a technical answer to a cognitive one. And, if anything, more agents means more potential interruptions.
For example, when you notice something mid-session. You have the full picture right now. Switch, and you'll forget where you are. If you don't, you'll forget what you discovered.
Staying focused is still how the best work gets done. The engineers doing the most important work aren't the ones juggling the most threads.
I've created a /sidequest skill to handle this. It pulls relevant context from the existing context window, enriches it, and creates a GitHub Issue with everything needed to work on it. I can immediately forget about it and move on.
But, here's the thing...
The new issue has everything it needs.
Why wait?
My background agent picks up these issues and works on them in a sandboxed environment. If it can't complete the task, I'll have more context and a note on the issue. But if it completes the task, I'll have a draft pull request to review.
All without leaving the task I was working on.