The Agentic Digital Twin-Led Assessment System (ATLAS): A new architecture for psychological assessment in the era of agentic AI

Psychological assessment has spent a century refining a single architecture. A person sits down, to complete an instrument designed for everyone, and receives a score positioned against a population mean. Each technological wave has improved this architecture without replacing it. But digital twins, agentic AI, and swarm intelligence now make it possible to specify a different assessment architecture: one in which the model of the person is built first, the assessment is generated on demand from that model, and evaluation preserves disagreement as diagnostic signal rather than averaging it away. This paper introduces the Agentic Digital Twin-Led Assessment System as a five-layer conceptual architecture coupled by a continuous feedback loop, with the AI-IARA framework functioning as a set of engineering-grade design constraints. This shifts psychology’s role from validating instruments after they are built to specifying assessment infrastructures that protect validity, and agency. ATLAS offers a conceptual architecture for precision psychology.