Enzyme engineering and discovery are crucial for a future sustainable bioeconomy. Harvesting new biocatalysts from large libraries through directed evolution or functional metagenomics requires accessible, rapid assays. Ultra-high throughput screening formats often require optical readouts, leading to the use of model substrates that may misreport target activity and necessitate bespoke synthesis. This is a particular challenge when screening glycosyl hydrolases, which leverage molecular recognition beyond the target glycosidic bond, so that complex chemical synthesis would have to be deployed to build a fluoro- or chromogenic substrate. In contrast, coupled assays represent a modular plug-and-play system: any enzyme-substrate pairing can be investigated, provided the reaction can produce a common intermediate which links the catalytic reaction to a detection cascade readout. Here, we establish a detection cascade producing a fluorescent readout in response to NAD(P)H via glutathione reductase and a subsequent thiol-mediated uncaging reaction, with a low nanomolar detection limit in plates. Further scaling down to microfluidic droplet screening is possible: the fluorophore is leakage-free and we report a three orders of magnitude improved sensitivity compared to absorbance-based systems, so that less than one turnover per enzyme molecule expressed from a single cell is detectable. Our approach enables the use of non-fluorogenic substrates in droplet-based enrichments, with applicability in screening for glycosyl hydrolases and imine reductases (IREDs). To demonstrate the assay's readiness for combinatorial experiments, one round of directed evolution was performed to select a glycosidase processing a natural substrate, beechwood xylan, with improved kinetic parameters from a pool of >10e6 mutagenized sequences.