Hacker News Reader

I've got a light version of this with local models; just one coding agent, one task takes 1-5 minutes. All local on constrained hardware helps; can't really run a ton of agents in parallel at good speeds.

During each task I context switch to some other work, emails, chores etc.

Important to take breaks and assess before starting a new session.

View on HN

I can do this somedays. Others I find myself staring at the reply panel in an email with my mind still remotely focused on the potential outcome of my last prompt. I think is what gets me. In theory, I should be able to both. But in reality: I’m 1.5x tasking: not entirely multi tasking when in comes to context shifts.

Reply on news.ycombinator.com