Using autonomous molecular analytical devices embedded within an ocean observatory, we studied harmful algal bloom (HAB) ecology in the dynamic coastal waters of Monterey Bay, California. During studies in 2007 and 2008, HAB species abundance and toxin concentrations were quantified periodically at two locations by Environmental Sample Processor (ESP) robotic biochemistry systems. Concurrently, environmental variability and processes were characterized by sensors co‐located with ESP network nodes, regional ocean moorings, autonomous underwater vehicle surveys, and satellite remote sensing. The two locations differed in their long‐term average physical and biological conditions and in their degree of exposure to episodic wind‐forced variability. While anomalously weak upwelling and strong stratification during the 2007 study favored toxigenic dinoflagellates (Alexandrium catenella), anomalously strong upwelling during the 2008 study favored toxigenic diatoms (Pseudo‐nitzschia spp.). During both studies, raphidophytes (Heterosigma akashiwo) were detected within a similar range of concentrations, and they reached higher abundances at the relatively sheltered, stratified site. During 2008, cellular domoic acid reached higher concentrations and was far more variable at the shallower ESP node, where phytoplankton populations were influenced by resuspended sediments. Episodic variability caused by wind forcing, lateral mixing, internal waves, and subsurface phytoplankton layers influenced ESP detection patterns. The results illustrate the importance of mobilizing HAB detection on autonomous platforms that can intelligently target sample acquisition as a function of environmental conditions and biological patch encounter.