Agentic Workflows in Senior-Living Technology: Design Patterns, Failure Modes, and Trust Boundaries for AI and Voice Agents in Resident Services
Senior-living operators are deploying AI and voice agents to handle routine resident requests. The people using these agents are not typical software users: residents may have hearing loss, slower or atypical speech, cognitive decline, or a health problem that comes up in the middle of an ordinary request. This white paper argues that in care-adjacent settings, safety has to be built into the architecture rather than left to prompt wording. The paper contributes: a request taxonomy that classifies resident requests by what goes wrong if the agent gets them wrong a reference architecture with a two-stage safety router that runs before the language model and fails safe if it breaks eight design patterns, five of them mandatory, including read-back confirmation, a human always reachable, and escalation of any health request to a person fourteen failure modes specific to older users, each with mitigations an escalation policy that separates an acknowledged alert from a responder actually reaching the resident an audit model mapped to the NIST AI RMF an evaluation method with a ten-scenario simulation specification The framework builds on two systems OEAT has deployed: a pre-sales AI receptionist for senior-living communities, and an Alexa voice service that lets residents hear dining menus and sign up for activities and transport. It closes with a phased roadmap of further agents. The resident-services agent described is a proposed design; the paper reports no field or simulation outcomes.
Authors
- Diwesh Saxena (ORCID: https://orcid.org/0009-0001-9546-841X)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-14
- DOI
- https://doi.org/10.5281/zenodo.24217724
- Primary Topic
- Technology Use by Older Adults
- Type
- article
- Field-Weighted Citation Impact
- 0.00