Cheap design

Software has always had only one cost: design. Fabrication - compilation, copying - is essentially free. But because design meant expensive human labor, we converged on the same pattern as physical manufacturing: build everything from standard parts. Libraries, frameworks, package managers. You adapt your design to available parts, write glue code and adapters, accept someone else's interface and bugs, and deal with transitive dependency conflicts. The resulting system is larger and more fragile than ideal, but each part is well-tested and writing from scratch costs too much.

LLM-assisted coding collapsed the cost of design for an entire category of code. The migration script or data converter that wasn't worth twenty minutes to write by hand now takes twenty seconds to generate. But unlike 3D printing in the physical world, LLM-generated code has no quality penalty - it's the same bytes running on the same CPU. If it's correct, it's indistinguishable from hand-written code. This is as if 3D printing suddenly produced parts with the material properties of machined steel.

The dependency calculus flips completely. The old question was "is there a library for this?" The new question is "is this problem hard enough to justify taking on a dependency?" Cryptography, compression, HTTP clients - yes, these are genuinely hard problems with never-ending CVE streams. An ORM that imposes its own model to save you from SQL - maybe. A logging framework with pluggable backends when you need JSON to stderr - probably not. A configuration library with layered overrides and hot-reloading when you need five environment variables - no.

The result: codebases become smaller for the same functionality. Custom code that does exactly what you need is almost always shorter than a library import plus the glue, configuration, and adapters to make it fit. In the standard-parts world, "custom" meant reimplemented everything badly. In the cheap-design world, "custom" means one piece designed for this exact purpose. Less dependency management, less glue, less accidental complexity from adapting your problem to someone else's abstraction. Design got cheaper, not free - what matters more now is understanding the problem you're solving, not the mechanical knowledge of expressing the solution in code.