Improving our ability to build and test DNA sequences will accelerate progress in biology. Multiplexed functional assays (MFAs) can test thousands to millions of DNA sequences for biological function, illuminating comprehensive sequence-function relationships at base-pair resolution. Though transformative, MFAs are currently limited by the sequences that can be built. Natural sequences can be mutagenized, allowing for the generation of all single-amino acid mutants of a particular protein. However, mutagenesis can only explore small subsets of sequence space, far smaller than the typical distance between homologous proteins. Alternatively, small (<200nt) arbitrary DNA sequences can be synthesized as microarray-derived oligo pools for use in MFAs. Unfortunately, sequences over 200nt are difficult to synthesize on microarrays, preventing the generation of protein-length (300-3000nt) libraries. Gene synthesis from microarray-derived oligos is a promising solution to this problem, allowing for the isolated construction and assembly of long DNA sequences. Unfortunately, the current cost of synthesizing genes from microarray-derived oligos is prohibitive, limiting scalability. In this dissertation, I describe the development of improved methods for multiplexed gene synthesis from microarray-derived oligos. First, I demonstrate the accurate quantification of polymerase error rates and error correction methods in synthetic gene constructs using next-generation sequencing. Next, I describe DropSynth, a low-cost, multiplexed method which builds gene libraries by compartmentalizing and assembling microarray-derived oligos in vortexed emulsions. Finally, I optimize polymerase choice, add error correction, and increase scale to significantly improve the fidelity and scalability of DropSynth. Taken together, these developments represent a new paradigm for the synthetic construction of gene libraries.