The recent massive release of new, man-made substances into the environment requires bioremediation, but a very limited number of enzymes evolved in response are available. When environments have not encountered the potentially hazardous materials in their evolutionary history, existing enzymes have to be repurposed. The recruitment of accidental, typically low-level promiscuous activities provides a head start that, after gene duplication, can adapt and provide a selectable advantage. This evolutionary scenario raises the question whether it is possible to adaptively improve the low-level activity of enzymes recruited from non- (or only recently) contaminated environments quickly to the level of evolved bioremediators. Here we address the evolution of phosphotriesterases (enzymes for hydrolysis of organophosphate pesticides or chemical warfare agents) in such a scenario: In a previous functional metagenomics screening we had identified a promiscuous phosphotriesterase activity of the α/β-hydrolase P91, with an unexpected Cys-His-Asp catalytic triad as the active site motif. We now probe evolvability of P91 using ultrahigh-throughput screening in microfluidic droplets, and test for the first time whether the unique catalytic motif of a cysteine-containing triad can adapt to achieve rates that rival existing phosphotriesterases. These mechanistically distinct enzymes achieve their high rates based on catalysis involving a metal-ion cofactor. A focussed, combinatorial library of P91 (> 105 members) was screened on-chip in microfluidic droplets by quantification of the reaction product, fluorescein. Within only two rounds of evolution P91's phosphotriesterase activity was increased ≈ 400-fold to a kcat/KM of ≈ 106 M-1s-1, matching the catalytic efficiencies of naturally evolved metal-dependent phosphotriesterases. In contrast to its homologue acetylcholinesterase that suffers suicide inhibition, P91 shows fast de-phosphorylation rates and is rate-limited by the formation of the covalent adduct rather than by its hydrolysis. Our analysis highlights how the combination of focussed, combinatorial libraries with the ultrahigh throughput of droplet microfluidics can be leveraged to identify and enhance mechanistic strategies that have not reached high efficiency in Nature, resulting in alternative reagents with a novel catalytic machinery.