A software engineer built a PC with 64GB of VRAM to run local AI models, rejecting subscriptions to OpenAI and Anthropic due to concerns about dependency on proprietary services. They previously found limited usefulness in local models but now observe that these models significantly improve the quality and speed of their software implementation. The post details their experience and motivations for the build.