Because using cheap LLMs to spam GH is way less expensive than fixing software with SOTA models.