AI was supposed to centralize power. Instead, two-person startups are shipping what used to take fifty people.
Not prototypes—real products. Customer support that works. Backend code that scales. Marketing that converts. The tools that once required teams now require taste.
The Open Source Acceleration
Open source models accelerated this. Llama, Mistral, Stable Diffusion—suddenly you don't need OpenAI's permission or pricing. Download the weights, fine-tune on your laptop, deploy on a potato. The moat became a bridge.
Big companies move slower with AI, not faster. Compliance departments, risk committees, enterprise procurement. Meanwhile, kids in dorm rooms ship AI features before the Fortune 500 finishes reading their AI generated "AI strategy deck."
Fine-tuning was ML expertise. Now it's a tutorial. Deployment was DevOps teams. Now it's a button. The barrier isn't technical—it's imagination and taste.
Small teams build different. Big companies use AI to optimize existing processes. Small teams use it to skip processes entirely. Why hire customer support when AI handles 90% of tickets? Why build a data pipeline when Claude can query your database?
The centralization thesis assumed AI would be like cloud—economies of scale favor giants. But AI is more like open source—once the knowledge exists, everyone has it.
David doesn't need to beat Goliath anymore. David builds around him. The giant's size isn't an advantage—it's organizational debt.