The 40 minute laptop hunt

A story about buying a used laptop the way we stopped buying things in 2016, and what happened when I let a browser agent run the search for me.

The machine I wanted

I run Omarchy, an Arch Linux setup with the Hyprland window manager, on hardware that must be x86_64. My MacBook is Apple silicon, so it cannot run the distro natively. Emulation is the only path and it delivers no GPU acceleration, which is fatal for a compositor. I needed a 2017 era laptop with 32GB of RAM, bought used, in Calgary, under C$550, and it had to be factory 32GB because DDR4 upgrade kits cost more than the machines.

The old way

The old way went like this. Open Kijiji. Scroll. Open Facebook Marketplace. Scroll again. Squint at a grainy photo of a palm rest. Drive across town, because a listing said Calgary, and find a machine worse than the listing, with a battery that lasts forty minutes and a warranty that does not exist. An evening of that buys you one candidate and a headache.

The market has no API. Kijiji, Facebook Marketplace and eBay all block plain fetch requests. A program that reads prices needs a real browser, logged in, tab by tab.

The hunt

I handed the search to Aside, the browser that runs itself. It opened real tabs and ran four passes in parallel. Price research across the model family. Refurb liquidator stores. eBay listings read from detail pages. Calgary classifieds on both platforms. Then it re-opened every top candidate in a fresh tab instead of trusting its own scrape, because scrapes lie.

The lies showed up fast. The best store listing promised a one year warranty. The live page said 30 days. The CPU model on the page disagreed with the URL, and a buyer review disagreed with both. A classified ad posted a 2GB graphics card as 4GB. In this market a claim is a question to verify, never a fact.

The surprise

The hardware math upended my assumptions. In PassMark, the older i7-6820HQ out-scores the newer i7-7820HQ because it holds turbo longer under load. The Xeon badge adds zero idle cost, both are 45W parts. With the dedicated GPU off, these workstations idle around 12 to 18 watts, which is nothing.

Then I changed one constraint mid-hunt. The laptop has to last a few hours on battery. That single sentence reshuffled the whole ranking, because every candidate is a 45W machine with a tired old pack, and a 28W 11th gen chip with modern idle states suddenly beat everything on value.

The receipt

Forty minutes of work produced raw listing links, a benchmark of each candidate against my own Mac, a PassMark per CAD ranking, power draw figures, and a battery floor. The machine I emailed a seller about was listed at C$230. I made that email myself. The agent screens, ranks and verifies, the human keeps the final call, and that division of labor is the point.

The recipe

  1. Filter by factory RAM. No upgrade projects, DDR4 kits cost more than the laptops.
  2. Verify every listing live in a browser tab. Read the detail page, trust nothing from snippets.
  3. Rank by PassMark per CAD dollar, then re-rank the moment a constraint changes.
  4. Ask the seller for battery health before deciding, that number flips the pick.
  5. The human emails the seller and clicks buy. The agent screens.