What separates an AI agent that becomes part of everyday operations from one that remains trapped inside an impressive demonstration?
In this episode of AI at Work, I speak with Jacob Kosior, who leads client strategy at EliseAI. The company builds vertical AI agents for the housing industry, handling property management workflows such as answering leasing inquiries, scheduling tours, processing renewals, collecting rent, and coordinating maintenance.
EliseAI says its technology is live across over six million housing units in the United States and Canada. Jacob brings an unusual perspective because he spent over a decade working in multifamily housing operations and was previously an EliseAI customer. He has experienced these systems from both sides of the relationship and works regularly with the operators using them.
We discuss why the agentic AI debate often becomes trapped between exaggerated expectations and deep skepticism. Some people believe agents can already perform almost any task, while others see them as chatbots with a new label. Jacob describes a narrower and far more useful reality: agents completing repetitive workflows from start to finish, provided they have access to the right systems, operational context, and escalation routes.
Housing provides several valuable examples. A conversation about unpaid rent may reveal that a resident is withholding payment because of an unresolved maintenance problem. Handling the complete situation requires an agent that can understand both workflows and connect the relevant information. EliseAI says the experience behind its agents includes over one billion conversations, helping the system account for edge cases it has previously encountered.
Jacob also discusses what separates production deployments from AI pilots that never progress. Adding a chatbot to an existing technology stack may answer basic questions, but it rarely changes how work gets done. An operational agent needs access to the systems, data, and context required to resolve a problem. It must also recognize when it has reached the limit of its ability and pass the customer to the person best equipped to help.
One of the most interesting lessons concerns AI acceptance. According to Jacob, residents generally prioritize a fast, accurate resolution over whether the response comes from a person or an AI agent. EliseAI also found that introducing familiar regional voices to its voice AI increased conversations and conversions. This suggests acceptance can depend on familiarity, responsiveness, and outcomes rather than the technology label.
We also consider how leaders can choose suitable workflows, why agents should be tested with difficult customer questions, and how automation could support heavily manual areas such as affordable housing administration.
Is your business testing whether an AI agent can sound intelligent, or whether it can genuinely resolve the customer’s problem? Listen to the conversation and share your thoughts with me.