The philosopher’s stone

Every day, another video appears telling us that we’re about to assemble an army of agents that will write code, build products, think for us, create things, and run companies. Plenty of people find this impressive.

Code has gotten dramatically cheaper. Prototypes come together faster. AI is already helping to build the next generation of AI systems. But there’s one unpleasant catch. If the person working with agents has no architectural vision, it can feel as though they’re building a palace with AI’s help. More often than not, though, what emerges is merely a very elaborate shed.

AI can boost productivity enormously, but it doesn’t replace taste, accountability, architectural vision, or an understanding of what should be built, why it should exist, what problem it solves, and which trade-offs you’ll inevitably have to accept along the way.

Real engineering is less about “writing code faster” and more about “seeing the constraints and points of failure, telling durable architecture apart from an imitation of it, and knowing what will improve reliability and what will only increase the scale of your problems.”