Comment by zozbot234

You're talking about multi-session vs. single-session throughput. A single user can easily leverage multiple sessions via e.g. subagent swarms, especially on a lower-end setup where any single session is going to be quite slow. Saturating utilization during off-hours is harder but potentially quite feasible by assigning lower priority, unattended tasks/inference loops.