AI is reaching for AGI, but right now it’s taking the scenic route. OpenAI’s newly released GPT-5 is billed as a “significant step toward AGI,” but it’s received criticism that under the hood it’s just clever repackaging. The model’s headline improvements include better code generation, fewer hallucinations, improved multi-step reasoning, and less sycophancy. This innovation is powered largely by a new router that directs queries to the most suitable existing GPT models, including a “Thinking” variant, so the novelty is not in groundbreaking algorithms or fresh data, like many might have expected. While this routing system could make GPT-5 more efficient, reliable, and cheaper to use, it also might hint at a plateau in LLM progress, with gains now coming from orchestration rather than raw capability jumps. Some see this as a pragmatic shift toward controllable, modular AI, while others interpret it as admission that current LLM architectures may have hit their limits on the road to true AGI. LLMs might not know everything, but delegation is a good start.