Translation note: This English version follows the structure, data, figures, and references of the corresponding Chinese post. It was reviewed and synchronized on 2026-08-03.

A Cyber Cleanup: Reinstalling an Old Laptop into an AI Agent Machine with Gemini

Project notes · Written on March 23, 2026, 4:00–5:40 PM

Background

OpenClaw had just gone viral, and I had experimented with it on a MacBook. But given the security concerns, token cost, and limited production readiness, I stopped there.

When I later cleaned up my room, I found an old laptop from my undergraduate days collecting dust. During a conversation with Gemini, I started thinking about turning it into a lightweight engineering machine or server.

This machine is only meant for lightweight or experimental AI tasks. If you need heavy development, you will need better hardware.

The goal was simple: keep no private files, install the lightest practical system, and dedicate the machine to AI coding execution.

Why not buy a server?

Save money
Reuse old hardware
Keep it fully private
Learn by rebuilding

The rule was simple: delete everything and start again. The rebuilt machine is only suitable for lightweight or experimental AI tasks; serious development needs better hardware.

Preparation

A stable internet connection
An 8GB+ USB drive
A large backup disk

Useful software:

Gemini / Doubao / Qwen web apps
Snipaste
Antigravity or VS Code + Cline
Motrix for P2P downloads
Rufus for writing system images to USB
Clash Verge
LocalSend for cross-platform local transfer
Chrome

The Process

Cleaning up files

I first cleaned up the machine’s files. This part is different for every computer, but the Downloads and Documents folders on the system drive deserve special attention. A tree listing from PowerShell or Bash can be shared with an AI tool so that potentially valuable documents can be identified and preserved.

I also used WizTree to visualize disk usage and speed up the cleanup.

I originally intended only to remove old software, not to reinstall Windows. Using Gemini through Antigravity, I asked “Why is my computer so slow?” and had the agent read the machine’s configuration and running processes.

It produced a Markdown hardware profile and noted that the computer had not been restarted for 248 days. Fragmented files and accumulated software were contributing to the slowdown.

The hardware was:

Intel Core i7-6500U at 2.50 GHz (2.60 GHz reported)
8 GB RAM (7.87 GB usable)
NVIDIA GeForce 940M (2 GB) and Intel HD Graphics 520 (128 MB)
256 GB LITEON CV1-CC256 SSD

At its worst, opening a browser took five seconds. I opened Task Manager with Ctrl+Shift+Esc and the installed-programs panel with Win+R → appwiz.cpl, captured screenshots with Snipaste, and asked Gemini to explain every process and program and produce an uninstall list. This uncovered vendor leftovers, bundled software, and other resource-heavy preinstalls.

Removing outdated and unnecessary software freed approximately 40 GB. The process taught me a lot about Windows permissions and deep cleanup, but some unwanted programs were so difficult to remove that I eventually decided to reinstall the system.

Downloading the system image

After discussing my requirements with Gemini, I chose Windows 10 Enterprise LTSC 2021 over Ubuntu 24.04 because of software compatibility and ecosystem support.

I downloaded the Windows 10 Enterprise LTSC 2021 (x64, Simplified Chinese) ISO via BT from next.itellyou.cn and used Motrix for P2P transfer. After finishing the download, I continued seeding until the share ratio reached 1 as a small contribution to the anonymous peer-to-peer network.

Making a bootable USB

I opened Rufus, selected the ISO, verified its hash, and then wrote it to a USB drive.

The checksum matched, so the image had not been tampered with or poisoned.

I then started the USB-writing process and selected every option except “use the current user’s regional settings.”

Reinstalling the system

After one final check that all files had been backed up or removed, I asked AI to generate a Markdown SOP for the reinstall process. I transferred it to another device—or photographed it with a phone—so that it could be followed while the laptop was offline.

I configured the BIOS to boot from USB and installed Windows LTSC. SSDs did not need partitioning, while mechanical disks were partitioned normally. If something is unclear, a photo can be sent to Gemini or Doubao for guidance; careful operation should avoid turning the machine into a large paperweight. If it does become a paperweight, please do not blame the author.

After installation, I let Windows Update patch the system, which took about 30–60 minutes.

Installing the base tools

I installed:

7-Zip
Windows Update Blocker
Defender Control
Clash Verge
ToDesk

Then it was time to configure OpenClaw / Claude Code / other AI tools.

What if an agent crashes the machine?

There is still a recovery route: return to the start of the article and follow the procedure again. After installation, Windows Update can patch security issues and drivers from 2021 onward; this usually takes around 30–60 minutes.

Epilogue: the cyber afterlife of an old laptop

As of 2026-03-23, I checked memory prices from two years earlier and found that they had surged from 900 RMB to 2,400 RMB. That is a good reminder of how much AI is changing hardware demand.

More and more document-heavy and decision-heavy jobs are being reshaped by agents. Whether AI is a bubble or a real transformation, the safe answer is still the same: keep up with the wave, or get left behind.

In other words: there was no shortcut, so I had to learn the whole thing the hard way.