Why should we assume that LLMs would be stagnant at current levels when, so far, they haven't stopped improving? I remember when simply using Copilot to just auto-complete 1 line of code was a groundbreaking unimaginable advancement. That was only a few years ago. New and improved models are being released nearly every week. The open-source models are nearly as advanced as the closed source models (if you have the hardware to run them at full capacity).