Source: a16z
The article identifies a gap in AI deployment: language models dominate production systems, but the economic value accumulates in automating physical tasks—manufacturing, logistics, robotics—where text-based AI has limited application. This requires different architectures (multimodal systems, embodied learning, real-time reasoning) and opens space for new vendors to compete outside the LLM moat OpenAI and Anthropic have built. Companies that crack reliable physical-world automation will capture far more industrial spending than those optimizing chatbot inference speeds.