Keep work running after the laptop closes

  • Moving execution off the daily laptop removes the hesitation to start work before a meeting, leaving the office, or sleeping. 59:05
  • His proposed home setup is an old computer running Ubuntu, connected through Tailscale, hosting the proxy on a residential connection, and available through T3 Code. 59:29
  • For a free starting machine, he suggests asking friends or family for an old laptop or desktop with at least 8 GB RAM and four cores. Later he suggests even 4–8 GB can be useful; this is workload-dependent advice. 59:50 1:02:25
  • His cost example compares a $275/month Hetzner server with 32 GB RAM and 16 threads against a $700 purchased box with 32 GB RAM and a 1 TB drive. He favors buying, at roughly two and a half months of the quoted rent, rather than claiming the rented machine is the bargain. 1:00:11 1:00:29
  • These boxes execute tools and builds; they do not run model inference locally. He estimates 60–80 W at worst for the discussed machine and expects only one or two CPU threads busy most of the time. 1:00:55 1:01:15
  • His 32-core server shows CPU usage rounding to 0% with about six agent threads active. This snapshot supports his claim that waiting agents alone need little CPU; it is not a sustained build benchmark. 1:01:15 1:01:38
  • He prefers Linux and attributes Mac parallel-work limits to filesystem and security behavior. This is his diagnosis and experience, not an operating-system benchmark demonstrated here. 1:01:53 1:02:25