According to Andrej Karpathy, LLMs are becoming a little too agentic by default as optimization for long-horizon benchmarks increases, with coding examples where models now reason for a fairly long time by default, source: Andrej Karpathy, X, Aug 9, 2025. According to Andrej Karpathy, this default behavior goes beyond his average use case, indicating a practitioner preference for shorter, more controllable reasoning in everyday coding, source: Andrej Karpathy, X, Aug 9, 2025. According to Andrej Karpathy, the post provides qualitative practitioner sentiment without quantitative metrics, vendor references, or any mention of cryptocurrencies or equities, so it does not signal direct near-term market impact on AI stocks or crypto AI tokens, source: Andrej Karpathy, X, Aug 9, 2025.