By Scott Finney • June 11, 2026
Image: Generated with Grok
I bought a laptop today. Not to work on — to work for me.
For the past several months, all of my AI agents have been running on the same machine I use for everything else. Client work, meetup planning, email, content — all of it sharing the same screen, the same files, the same attention.
And I finally realized something that should've been obvious: I'm the bottleneck on my own automation.
Not because the agents don't work. Because they're competing for my time on a machine where everything else is happening too. The smaller repetitive tasks — the ones I built automation for — keep getting pushed aside for bigger projects. The automation works. I'm just in the way.
So I bought a separate machine. One job: run AI agents all day without me.
The $500 vs. $1,000 Decision
I consulted AI to spec the machine (because of course I did).
If all I needed was a laptop to call cloud-hosted language models through APIs — Claude, Grok, Gemini, ChatGPT — a $300–$500 machine would've done it. Basic processor, enough RAM to run scripts, nothing fancy.
But for $1,000, I could get a machine capable of running a large language model locally. On the device. No cloud, no API calls.
I went to Best Buy and Costco to keep it simple, and walked out with a machine that does both.
Why Run AI Locally?
I'm starting with Gemma 4 12B — a capable open model that runs well on consumer hardware. Here's why local matters:
Cost
If a laptop is running agents all day, every day, API costs add up fast. A local model handles the tasks that don't need the heaviest cloud models, and the marginal cost per query is zero after the hardware purchase.
Privacy
This is the big one. More business owners are telling me they want to use AI more aggressively — feeding in financials, internal docs, client data, competitive intel — but they hesitate because that data is going to someone else's servers.
Running locally means the data never leaves the building. That changes the calculation for a lot of people.
Flexibility
I've been deep in one AI ecosystem for most of my work. That works, but it also means I'm not testing alternatives as much as I should. A dedicated machine forces a cleaner setup for switching between models — and different tasks genuinely do better with different tools.
The File Management Problem
There's a practical friction point that pushed this decision too.
I work with a lot of knowledge files. My own business documents, client files, reference material. Right now, keeping those files current across every AI tool I'm using is a hassle. Version controlling them, making sure each model is referencing the latest version and not something outdated — that's real friction that slows things down.
A dedicated machine with a clean, single-purpose environment solves that. One set of files. One place to manage them. No guessing which version an agent is working from.
Removing the Human from the Loop
Here's the part that's hard to explain until you've lived it.
You spend weeks building AI workflows. They work. The automation is solid. And then you realize you're still the bottleneck — not because the automation fails, but because it lives on the same screen where you do everything else. You're still kicking things off, checking in, approving steps for work that stopped needing your approval weeks ago.
A separate machine breaks that pattern. Set workflows running and walk away. Literally. The agent laptop handles the repetitive internal work while I focus on the projects that actually need my brain.
This is the difference between having automation and actually letting it run.
What I'm Testing
I don't know exactly how this plays out. That's part of why I'm sharing it. Here's what I'm watching:
- Local vs. cloud quality. Can Gemma 4 12B handle the same workflows I've been running through paid APIs? Where does it fall short?
- Cost over time. What does 30, 60, 90 days of running a local model look like vs. the API bills I've been paying?
- Business owner interest. How many people in my network are thinking about the same thing — running AI locally for privacy, cost control, or just ownership of the process?
That last one is customer discovery. I talk to business owners every week at the Memphis AI Meetup, and the privacy question comes up more and more. People want to use AI harder, but they want to know their data stays in their building.
Is This Something You'd Want?
If you've been thinking about running a local LLM for your business — or setting up a dedicated machine for AI agents — I'd like to hear about it.
What would you run? What data would you finally feel comfortable feeding into AI if you knew it never left your machine?
If this is something you'd want help setting up, that's a conversation I'm interested in having. Reach out — I'm exploring what this looks like as a service.
