Automated in vivo enzyme engineering accelerates biocatalyst optimization. The development of stable and selective biocatalysts is essential for cost-competitive bio-based processes. Traditional methods are low throughput and labor-intensive, but strategies like in vivo screening and evolution, combined with machine learning (ML), can enhance throughput and reduce manual labor. This article proposes an integrated solution using ML-guided automated workflows, including library generation, hypermutation systems, and in vivo growth-coupled selection, to accelerate biocatalyst optimization. Enzyme engineering is crucial for improving biocatalysts in various applications, including biopharmaceuticals, industrial biotechnology, and bioremediation. Current efforts aim to design biological systems with enzymatic activities beyond natural capabilities. Directed evolution, which allows Darwinian evolution in a test tube, is a key approach, with in vivo evolution being particularly promising. Automated biofoundries support high-throughput efforts in engineering biology, and AI/ML aid in designing new biological systems. The integration of ML, in vivo continuous evolution, and automated biofoundries can accelerate the generation of new and competitive biocatalysts. An integrated workflow combines ML for predicting enzyme modifications and designing selection strains, followed by in vivo hypermutators and growth-coupled selection. This process can be iterated for multiple rounds. De novo enzyme design complements enzyme engineering by creating novel enzymes from scratch. Tools like Rosetta predict and optimize enzyme sequences. ML supports de novo design by identifying relevant residues and predicting protein structures. De novo design can generate biocatalysts free from existing enzyme constraints. ML-supported pathway design increases the engineering design space by automating pathway design stages. It aids in predicting metabolic reactions and analyzing data for bioprocess optimization. The METIS workflow demonstrates efficient optimization with minimal experiments. Selection strains enable high-throughput in vivo enzyme screening by using auxotrophic or antimetabolite selection. These strains allow growth restoration through enzymatic activity, serving as a readout for module activity. Different selection strains can be used for varying sensitivity ranges. Combining growth-coupling with directed evolution enhances new phenotypes. In vivo mutagenesis strategies, such as MAGE and CRISPR-Cas, enable gene diversification within cells. Hypermutation methods increase mutation rates, facilitating the exploration of the fitness landscape. ALE further enhances emerging phenotypes by applying selective pressure through different cultivation conditions. It allows the selection of optimal variants and can be combined with directed evolution to explore rugged fitness landscapes. Non-canonical hosts like P. putida and V. natriegens are increasingly used for in vivo engineering due to their robustness and suitability for specific reactions. These hosts can be used for automated in vivo enzyme engineering, with potential for future applications in bioprocesses. The integration of ML, automation, in vivo mutagenesis, and growth-coupled selection in a biofoundry enables the directed evolution of enzymes. This approachAutomated in vivo enzyme engineering accelerates biocatalyst optimization. The development of stable and selective biocatalysts is essential for cost-competitive bio-based processes. Traditional methods are low throughput and labor-intensive, but strategies like in vivo screening and evolution, combined with machine learning (ML), can enhance throughput and reduce manual labor. This article proposes an integrated solution using ML-guided automated workflows, including library generation, hypermutation systems, and in vivo growth-coupled selection, to accelerate biocatalyst optimization. Enzyme engineering is crucial for improving biocatalysts in various applications, including biopharmaceuticals, industrial biotechnology, and bioremediation. Current efforts aim to design biological systems with enzymatic activities beyond natural capabilities. Directed evolution, which allows Darwinian evolution in a test tube, is a key approach, with in vivo evolution being particularly promising. Automated biofoundries support high-throughput efforts in engineering biology, and AI/ML aid in designing new biological systems. The integration of ML, in vivo continuous evolution, and automated biofoundries can accelerate the generation of new and competitive biocatalysts. An integrated workflow combines ML for predicting enzyme modifications and designing selection strains, followed by in vivo hypermutators and growth-coupled selection. This process can be iterated for multiple rounds. De novo enzyme design complements enzyme engineering by creating novel enzymes from scratch. Tools like Rosetta predict and optimize enzyme sequences. ML supports de novo design by identifying relevant residues and predicting protein structures. De novo design can generate biocatalysts free from existing enzyme constraints. ML-supported pathway design increases the engineering design space by automating pathway design stages. It aids in predicting metabolic reactions and analyzing data for bioprocess optimization. The METIS workflow demonstrates efficient optimization with minimal experiments. Selection strains enable high-throughput in vivo enzyme screening by using auxotrophic or antimetabolite selection. These strains allow growth restoration through enzymatic activity, serving as a readout for module activity. Different selection strains can be used for varying sensitivity ranges. Combining growth-coupling with directed evolution enhances new phenotypes. In vivo mutagenesis strategies, such as MAGE and CRISPR-Cas, enable gene diversification within cells. Hypermutation methods increase mutation rates, facilitating the exploration of the fitness landscape. ALE further enhances emerging phenotypes by applying selective pressure through different cultivation conditions. It allows the selection of optimal variants and can be combined with directed evolution to explore rugged fitness landscapes. Non-canonical hosts like P. putida and V. natriegens are increasingly used for in vivo engineering due to their robustness and suitability for specific reactions. These hosts can be used for automated in vivo enzyme engineering, with potential for future applications in bioprocesses. The integration of ML, automation, in vivo mutagenesis, and growth-coupled selection in a biofoundry enables the directed evolution of enzymes. This approach