Could a computer scientist build a brain? We frame this as an engineering problem: write a program that a single cell executes to build itself into a brain. The program must fit in a genome-sized instruction set of about 1 gigabyte and halt within a developmental time window of at most about a year.

The algorithm must scale: it should work for the 302 neurons of C. elegans, the 10^5 neurons of a fly, the 10^8 neurons of a mouse, and the 10^10 neurons of a human brain. A computer scientist with no knowledge of developmental biology quickly realizes why obvious strategies fail at scale: the genome is too small to store per-synapse wiring, and axons searching blindly for targets would take too long.

These algorithmic constraints drive the design toward solutions resembling developmental strategies organisms actually use. The genome contains roughly 6 x 10^9 bits of information, requiring sub-linear scaling. Total axon length and developmental time must also scale sub-linearly. The computer scientist evaluates three approaches: the identity approach, the guidepost approach, and the coordinate approach.

Modern machine learning does not yet know how to generate large functional networks from compact descriptions. Analyzing biological development may inspire new algorithms for artificial systems. Features of neural development can be grounded in computational necessity rather than contingency alone.