Gene expression can be highly heterogeneous in isogenic cell populations. An extreme type of heterogeneity is the so-called bistable or bimodal expression, whereby a cell can differentiate into two alternative expression states. Stochastic fluctuations of protein levels, also referred to as noise, provide the necessary source of heterogeneity that must be amplified by specific genetic circuits in order to obtain a bimodal response. A classical model of bimodal differentiation is the activation of genetic competence in Bacillus subtilis. The competence transcription factor ComK activates transcription of its own gene, and an intricate regulatory network controls the switch to competence and ensures its reversibility. However, it is noise in ComK expression that determines which cells activate the ComK autostimulatory loop and become competent for genetic transformation. Despite its important role in bimodal gene expression, noise remains difficult to investigate due to its inherent stochastic nature. We adapted an artificial autostimulatory loop that bypasses all known ComK regulators to screen for possible factors that affect noise. This led to the identification of a novel protein Kre (YkyB) that controls the bimodal regulation of ComK. Interestingly, Kre appears to modulate the induction of ComK by affecting the stability of comK mRNA. The protein influences the expression of many genes, however, Kre is only found in bacteria that contain a ComK homologue and, importantly, kre expression itself is downregulated by ComK. The evolutionary significance of this new feedback loop for the reduction of transcriptional noise in comK expression is discussed. Our findings show the importance of mRNA stability in bimodal regulation, a factor that requires more attention when studying and modelling this non-deterministic developmental mechanism.