Software is Eating AI (2017)

The post argues that before AI can "eat software" as Jensen Huang predicted, software must first "eat AI," and that this is already happening as industry-developed frameworks like TensorFlow, Keras, and PyTorch spread computation-graph abstractions and conventions through the AI community. It contends that academic incentives historically favored hasty benchmark-beating code over good software engineering, while industry has perfected scalable APIs that commoditize complex tasks. The author predicts that future "application layers" will obviate most of today's AI development tasks, with natural language becoming the best way for end users to interact with the stack.