Member-only story
Software moats, agent architectures, and the engineering that still holds value when the cost of building drops to almost zero.
This article looks at software defensibility in a world where AI can generate working code almost for free. We start with a simple question: if building software costs almost nothing, what still gives it value, and then work through the classic moats of data, brand, distribution, and expertise to see which ones hold and which ones leak. From there it gets practical: how latency budgets shape voice pipelines, why human preference is hard to encode, how spec-driven development and Model Context Protocol change the way we build, and what irreversibility means once an agent can touch a database or a motor.
By the end, you will have a clear map of where the value in building software has moved, and patterns and tools you can use to build on that new ground.
Reference here.
The premise: cost and value are falling together #
The premise is now a cliché in engineering circles, but it is worth stating cleanly. The cost of building software is approaching zero. Teams that once shipped ten or twenty variations for…