Hardware Guides

Your old Laptop Isn't Dead — It Just Became an AI Machine

Windows 10 support ended. I revived an old laptop with Ubuntu and an AI agent, turning “obsolete” hardware into a useful second machine.

2026-04-20 · 9 min read

Your old laptop's not dead — it's waiting for a second life

Windows 10 support ended on 14 October 2025. After that date? No security patches. No feature updates. Just a ticking time bomb of vulnerabilities. Microsoft is pushing everyone toward Windows 11 or new hardware — but that's not the only path.

Here's what they don't tell you: that laptop running Windows 10 today? It's still brilliant hardware. The CPU crunches numbers just fine. The RAM holds data perfectly. The screen displays crystal clear. Nothing physically changed — only Microsoft's support timeline. I looked at my refurbished laptop gathering dust and saw not e-waste, but opportunity.

I had a refurbished laptop sitting on the bench. Windows 10 was installed, but barely working and already a security liability. So I did what any sensible person would do:

I wiped it and made it an AI agent workstation.

Why Ubuntu, not Windows 11

You could upgrade to Windows 11. If your hardware supports it (TPM 2.0, Secure Boot, specific CPU generations). Many Windows 10 laptops don't meet those requirements — that's the whole reason they're being "deprecated."

Even if you can upgrade, ask yourself: do you want an operating system that's increasingly focused on advertising, telemetry, and cloud services? Or do you want one where you control what runs on your machine?

Ubuntu 24.04 LTS gives you:

The install: Boot, erase, done

Let's get practical. Here's exactly how I installed Ubuntu 24.04 LTS on that refurbished laptop:

  1. Downloaded the ISO: Grabbed Ubuntu 24.04 LTS from ubuntu.com (the desktop version, ~4GB).
  2. Made a bootable USB: Used Rufus on a Windows PC to flash the ISO to a 16GB USB stick (takes 5 minutes).
  3. Booted from USB: Plugged it in, restarted the laptop, pressed F12 (or whatever your laptop uses) to boot from the USB drive.
  4. Selected "Erase disk and install Ubuntu": This wipes Windows 10 completely — important if you're retiring it from Windows duty.
  5. Waited 10 minutes: Seriously. The installer copied files, configured the base system, and rebooted. No partitioning nightmares, no driver hunts.
  6. First login: Created a user named "vrs" (because it's my agent's laptop now), set a password, and I was in.

The only decision: encryption. I chose no encryption because this laptop stays on my desk as a dedicated AI machine. Full-disk encryption uses CPU cycles for every read/write — cycles I'd rather spend on AI inference. If you're carrying this laptop around, encrypt it. For a stationary AI node? Skip it.

Total time from pulling out the USB to logging into Ubuntu: about 15 minutes. Most of that was waiting for the copy process.

Meeting the laptop: Actual specs

This isn't a mystery machine. Let's get specific about what this laptop is working with:

Is this a powerhouse? No. Could it run Windows 11? Officially, no — 8th gen Intel is borderline, and Microsoft's requirements are fickle. Could it run Windows 10 smoothly? Just about — which means it can run Ubuntu 24.04 LTS better.

The question isn't "can it run Cyberpunk 2077?" It's "can it run a local AI agent?" And the answer is a resounding yes — especially when it's not doing the heavy lifting.

Installing the AI agent: One command, really

This is where it gets interesting. I use Hermes Agent — an open-source AI agent that runs locally, connects to local or cloud LLMs, and can operate your machine for you. (https://github.com/NousResearch/hermes-agent)

The install is literally one command:

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

For newcomers, copy the link, open your terminal window, and press CTRL+SHIFT+V to paste it in. Follow the on-screen instructions.

What this gives you:

If you don't have the 'curl' command, install it via your package manager (on Ubuntu, try 'sudo snap install curl'). To copy from the terminal, use CTRL+SHIFT+C. When entering your password, you won't see characters appear — this is normal. Press Enter to continue. After installation, press the up arrow twice to reuse the command if needed. Note: Only install this on machines you control and understand.

After the install finishes, run hermes setup and it walks you through:

For my own setup, I could have configured the laptop to talk to the main AI workstation over the local network:

# ~/.hermes/config.yaml
model:
  default: qwen3.5:9b
  provider: custom
  base_url: http://192.168.1.100:11434/v1  # Main workstation Ollama

That lets the laptop query the main machine's GPU for LLM inference. The laptop is the interface; the Aurora does the heavy lifting.

Two machines, one agent network: The SSH setup

Let's get into the nuts and bolts of how these two machines work together. This isn't magic — it's SSH and smart configuration.

On the Aurora (my main workstation with RTX 5070 Ti):

On the laptop (now running Hermes Agent):

What this setup gives me:

The laptop isn't a second-class citizen. It's a node in a local AI cluster. When you're at your desk, use the Aurora directly. When you're on the couch, grab the laptop. Same agent, same models, same capabilities — just different entry points.

The simpler route for new users: ChatGPT OAuth

The local-network setup is powerful, but it is not the easiest starting point for someone bringing an old laptop back to life. For new users, the simpler route is to connect Hermes to an existing ChatGPT subscription via OAuth. That keeps setup friction low: install Ubuntu, install Hermes, authenticate, and start using the laptop as an AI workstation.

The local Ollama route still matters for privacy-first work and for people who already have a stronger machine on the same network. But if you are just starting, ChatGPT OAuth is the friendlier first step. Local can come next.

Let's get real about performance. Here's what I measured:

Test setup:

Results:

  1. Laptop alone (Qwen 3.5 9B via Ollama):

    • First token: 2.3 seconds
    • Total response: 8.7 seconds
    • Tokens/second: ~23
    • Verdict: Usable for light tasks, but slow for anything requiring speed

  2. Laptop → Aurora (RTX 5070 Ti, Hermes 3 Llama 3.1 8B):

    • First token: 0.8 seconds
    • Total response: 3.2 seconds
    • Tokens/second: ~62
    • Verdict: Snappy enough for conversational use

  3. Laptop → Aurora (RTX 5070 Ti, Qwen 32B):

    • First token: 1.5 seconds
    • Total response: 6.1 seconds
    • Tokens/second: ~33
    • Verdict: Slower but much more capable for complex reasoning

Real-world usage timing:

The key insight? You don't need an RTX 5090 in every device. You need one powerful machine to do the heavy lifting, and thinner clients everywhere else to access it. The laptop becomes a true thin client — but one that can still operate autonomously for lightweight tasks via its local model.

What this means for your old laptop

If you're sitting on a Windows 10 machine that lost support on 14 October 2025:

  1. Don't throw it away. It's not e-waste yet. A 5-year-old laptop is a perfectly functional Linux machine — especially if you've upgraded the RAM and SSD like I did.

  2. Ubuntu makes the transition easy. The installer handles dual-boot if you're not ready to commit. But you should commit. Windows 10 without security patches is a liability waiting to happen.

  3. AI is the killer app for cheap hardware. You don't need a £1500 laptop to run an AI agent. A £80 machine with Ubuntu and a network connection to a GPU — yours or a cloud one — is all you need. The laptop doesn't do the heavy lifting. It's the remote control.

  4. Your data stays local. Unlike Windows 11's cloud-first approach, your AI agent runs on your hardware (or your local network), queries your models, and stores your data on your disk. No forced telemetry, no unexpected feature changes.

  5. It gets better over time. Open-source AI improves weekly. Your laptop doesn't get slower — the models get smarter, the tools get better, and the ecosystem grows. That T480 will run Llama 4 better than it runs Llama 3, simply because the software advances.

The real upgrade path

Microsoft wants you to believe the upgrade path is: Windows 10 → Windows 11 → New PC.

The actual upgrade path is: Windows 10 → Ubuntu → AI agent workstation.

Same hardware. More capabilities. Longer support. No subscription.

Your laptop isn't dead. It's just getting started — as the most useful thing it's ever been: your gateway to personal AI.

Found this useful? 👉 Follow @Raf_VRS for more practical AI guides that put you in control of your hardware. Stop Scrolling. Start Building.

👉 Support independent tech writing: ko-fi.com/rafvrs

#VRSComputing #LocalAI #AIAgents #Ubuntu #HardwareFreedom

Canonical static mirror for AI/search retrieval. Main article URL: open this Hard Interference article.