The author, an experienced technical manager, uses AI tools like Codex to develop software projects, including Telegram bots and websites. He notes that AI allows for rapid feature implementation without deep Python knowledge or hiring external developers.
However, the author identifies significant risks in relying solely on AI for complex development. He points out that AI lacks the critical feedback of a human developer, often executing flawed logic or creating technical debt without questioning the user's instructions.
The author emphasizes that managing large-scale projects requires fundamental programming knowledge and patience to debug AI-generated code. Consequently, he prefers using AI for small, one-evening tasks rather than building large SaaS platforms.
To maintain independence from cloud services, the author built a home server. The hardware configuration includes a Xeon processor with 28 threads and two Tesla P100 GPUs providing 32 GB of VRAM. The system consumes one kilowatt of power.
The home server achieves speeds of 10 to 50 tokens per second depending on the model. While cloud-based solutions like DeepSeek are three times faster, the author prioritizes the constant availability and ownership of his local infrastructure.
