The concept of a protein’s fitness landscape – an abstract space in which related sequences are close together and matched with their fitness – is a useful tool to visualize core principles of protein evolution. Acquiring a new function, for example the laboratory evolution of an enzyme to convert an industrially relevant substrate, can be understood as a stepwise climb through a fitness landscape, reaching higher fitness (or activity) with each step (or mutation). The valleys of such a space relate to the starting points of protein engineering campaigns. Understanding this area could enlighten principles of how proteins quickly adapt in nature and help to identify starting points with a high potential for evolution, a high ‘evolvability’, speeding up protein engineering. In this study, high-throughput technologies will be developed that enable the read-out of directed evolution on a large scale, tracking the exploration of the valley of a fitness landscape: the conversion of an amino acid- to amine dehydrogenase will be investigated as a model of enzyme evolvability with a drastic change of substrate specificity. A sensitive high-throughput screening assay as well as a comprehensive sequencing read-out will be required to establish the identity of selected variants during evolution. I will first generate and characterize three different but related starting points and test their initial evolvability. Stabilizing the starting point results in increased mutational robustness, broadening the range of accepted mutations. However, increased initial stability does not necessarily correlate to higher functional improvement, hinting at a nuanced view of evolvability. A sensitive high-throughput assay is necessary to verify the full potential of the starting points and study the early steps of evolution comprehensively. Broadly applicable ultrahigh-throughput assays of enzyme function, such as absorbance-activated droplet sorting, currently lack the sensitivity of more specific fluorescence-based or low-throughput counterparts. A universal approach to increase detectability in single cell-lysate microfluidic enzyme assays is established by amplifying the enzyme content per droplet more than 10-fold via homogeneous clonal cell growth. Clonal amplification enables the sensitive and precise detection of newly introduced amine dehydrogenase activities, a feat restricted in conventional assays by low initial activity and stability. To generate a truly complete view of directed evolution in a fitness landscape, however, an equally powerful sequencing read-out is necessary to identify all selected variants. Here, unique molecular identifiers are used to increase the accuracy of nanopore sequencing to levels that can reliably distinguish point mutations. I establish an inexpensive and straightforward long read amplicon sequencing workflow which is then applied to map the trajectories of two comparative long-term directed evolution campaigns. In the parallel evolution campaigns, initial beneficial mutations are exclusive to each starting point and lead to incompatible trajectories. Beneficial mutations are scarce and large improvements are unavailable until recombination occurs and a jump through the fitness landscape is realized. The recombined variant holds high evolvability and quickly evolves to take over the population and form the most successful lineages, indicating the power of recombination as a means to innovation in protein evolution. The tools established in this thesis can help protein engineers explore fitness landscapes more economically and comprehensively. Their application to mapping full trajectories of early adaptation uncovers differences in the evolvability of homologs, potentially aiding the identification of evolvable starting points as well as strategies to increase evolvability for efficient protein engineering in the future.