Your AI agents are roommates: before buying a more powerful computer, measure what they leave running

By Nate (Nate's Substack) · October 20, 2026.

The author argues that you are no longer the only user of your computer. If you run agents throughout the day, they may open more browsers, launch more processes and write more files than you do. He says that, time and again, when he opened just one more program as a human, the system asked him to close something, and that he suffered an out-of-memory crash while working against a deadline. Everything seemed fine until it wasn't.

According to Nate, you are still the owner of the machine and the one who decides what matters, but you share its resources with software that does far more than answer questions in a chat. Agents have become "roommates," and almost nobody has set the house rules.

That explains part of the appeal of buying a bigger machine (a Mac mini, a Studio, an NVIDIA box): more room sounds great when the current situation is painful. Sometimes more room is exactly what the work needs. But, he argues, a faster chip won't stop a development server from staying open after a project ends, more memory won't decide which of twelve document-processing jobs should run first, and a bigger disk will eventually fill up if every experiment keeps its own environment forever.

His advice is that, before spending money, you should find out what agents leave running, what they leave behind and how much of that activity contributes to work you still want to do, and buy only for the remaining demand.

According to the preview, the article includes: (1) a check you can run today, using a read-only prompt that detects what agents leave running and what they leave behind, and reports without changing anything; (2) a "stop signal" for each agent, so that every task ends with its output saved somewhere findable and its leftover servers shut down; (3) a schedule for heavy work: start with one heavy task, then try two, and keep whichever option finishes sooner without making the machine uncomfortable to use; (4) a second use for the old computer, which can run the background queue so the new one stays usable, with Tailscale to reach it when you're away; and (5) a companion guide with the three full prompts, detailed hygiene instructions and setup tutorials for Ollama, LM Studio, Docker Desktop and Tailscale.

It also promises three worked-through decisions on keeping, reconfiguring, relocating or replacing a machine, and a method for comparing the first-year cost of each option.

The email is only a preview: the full content, the guide, the MCP for Nate's library and his Slack community are for paid subscribers, so no further details or figures from the full piece are available.