Auditory scene analysis (ASA) aims to retrieve information from the acoustic environment, carrying out three main tasks: sound source localization, separation, and classification. These tasks are traditionally executed with a linear data flow: first, sound sources are located; then, each source is separated into its own audio stream; then, information pertaining to the source is extracted from each stream. Doing so makes the last tasks (separation and classification) highly sensitive to errors of the first task (localization), and results in a high overall response time. Recently, deep-learning-based artificial intelligence (AI) models have been employed to avoid this issue by developing very robust techniques to solve each main task. However, doing so increases their complexity which, in turn, results in an ASA system that is non-viable in applications with restricted computational resources and that require low response times, such as bioacoustics, hearing-aid design, search and rescue, human–robot interaction, etc. In this work, a multi-agent approach is proposed to carry out ASA with feedback loops between techniques, such as: using the quality of the separation output to correct localization errors; and using the classification result to reduce the localization’s sensitivity towards interferences. Since less complex techniques can be used with this approach, the result is a multi-agent auditory scene analysis (MASA) system that bares a low overall response time. The proposed system was evaluated in several acoustic settings, and it showed to be robust against different types of noise, interferences and reverberation levels.