The user relies on frontier models like Codex and Claude for extensive tasks including a custom ERP system, FPGA test benches, DSP algorithm QA, and PCB firmware development, noting that usage costs are rising. They question whether a locally run LLM could replace these expensive services regardless of hardware expense.

Read original