Dogs are often considered as humans' oldest friends. The relationship between people and dogs (wherein they are regarded as family members in the home) is well-studied. Since social agents have recently become a part of family life, a new field of research has been introduced that is centered on the human-agent and dog-agent relationships. Hence, in this study, to investigate how social agents are perceived by dog owners and dogs, we employed existing communication agents such as smart speakers and conducted two experiments; the first one investigated dog owners' impression toward the agents in agent-owner interaction and the second one investigated dog's behavior in front of the agents during owner's speaking to the agents and the agents' speaking to the dogs. Furthermore, we statistically analyzed the results of thirty-three dog-owners and twenty-one dogs. The results obtained from the GODSPEED questionnaire analysis indicate that regarding aspects such as anthropomorphism, animacy, likability, and perceived intelligence, the dog-like smart speaker is significantly worse than the educational humanoid robots NAO and Pepper. Regarding aspects such as animacy, likability, and perceived intelligence, the dog-like smart speaker has a significantly lower rating than Google Home, and it has a lower rating than Pepper with respect to perceived safety. However, in the case of canine behavioral analysis, the percentage of individuals who engaged in social behaviors (such as sniffing the rear) in response to the dog-like smart speaker is found to be significantly higher than that for the other agents. Hence, although this study is an exploratory experiment in terms of subject limitations and determination of behavioral indicators, it is the first step towards bridging the gap between owners' and dogs' perceptions of an agent's sociality while employing the same physical agents. CCS CONCEPTS • Human-centered computing → Empirical studies in HCI; Empirical studies in interaction design; HCI design and evaluation methods.