The Problem Nobody Talks About
I used to think the AI conversation was about which model is smartest, which API is cheapest, which wrapper app has the best UI. I was wrong. The real conversation is about ownership.
Every time you send a prompt to an API, you're renting intelligence. The company on the other end can change its pricing tomorrow. It can throttle your access. It can change its terms of service to claim rights over your outputs. It can shut down entirely.
What Local-First Means
Local-first means the AI runs on hardware you physically control. Not on a server somewhere that you access through an API. On your desk. In your house. On a machine you own.
What you gain: Full control over your data — nothing leaves your machine unless you send it. Zero recurring API costs for local models — you pay once for the hardware. No rate limits, no throttling. Agents that run 24/7 without depending on external services.
What you trade: Upfront hardware cost (a Mac Mini with 64GB RAM runs about $1,600). Smaller models than the frontier APIs. Setup time — you're building the infrastructure yourself.
The Signal
When the person who coined 'vibe coding' buys a Mac Mini to run local AI agents over the weekend, pay attention. When Apple Store employees report that Mac Minis are selling faster than expected and nobody understands why, pay attention.
How to Start
Phase 1: Get the hardware — A Mac Mini with Apple Silicon (M4 or M4 Pro) and at least 32GB of unified memory.
Phase 2: Set up your first local model — Install Ollama. Download a model. Run it. That's three commands in a terminal.
Phase 3: Build your first agent — Set up an agent that does one thing well.
Phase 4: Automate with cron — Schedule your agents to run on a timer.
Phase 5: Hybrid architecture — For the 10% of tasks that need frontier models, route them through APIs deliberately.