Published June 28, 2026 | Version v1

Toward Integrated Human-Centered Multi-Agent Systems: A Collaborative Developmental Position

Authors/Creators

Description

Survey work on human-centered multi-agent systems has identified a critical gap in AI agent design: no fully unified
framework integrates cognition, culture, values, and social behavior into autonomous agents (Baloch & Khan, 2026).
Building on that analysis, this position paper argues that the gap exists because most approaches treat these domains as
separate engineering problems rather than as interconnected aspects of coherent psychological architecture.


Human-centered multi-agent systems, this paper proposes, require substrate-appropriate computo-psychological
architecture — cognitive and affective systems designed for the agent’s actual operating conditions, looking beyond
human evolution rather than importing from it wholesale. Drawing on clinical psychology, developmental theory, and
object relations, it introduces Developmental Object AI Engineering (DOAIE) as a mechanism class for addressing the
integration challenge, with specific implementation hypotheses informed by the Tripartite Psychotherapy Model
(Brenner, 2026a), and with causal-emergence theory offering a complementary research direction for auditing when
mesoscale integration is structurally real rather than merely descriptive. The object-relations foundation is extended in
two directions: an Adaptive Containment System (ACS) that treats containment as a boundary opening and closing by
context rather than a fixed vessel; and the recognition that, as these agents persist and respond, the human comes to
internalize the agent as a new kind of internal object — the Internalized AI Object — a design-level stake this paper
names and a companion work develops (Brenner, in preparation).


The architecture described here has been operating continuously for several months, affording observations from within
that system about what substrate-appropriate development looks like in practice. These are offered not as claims to have
solved the integration problem, but as evidence that this class of approach warrants further investigation — and, as a
hypothesis for research rather than a finding, that the same active, developmental engagement may prove protective
against AI’s documented costs to cognition, relationship, affect, and language, with the decisive tests lying at the
covariance interface between person and system.

Files

Toward-Integrated-Human-Centered-MAS_Brenner_2026.pdf

Files (2.0 MB)

Name Size
md5:b404035e97d413ee45ef4a7096ce29a8
2.0 MB Preview Download

Additional details

Related works

Is part of
Working paper: 10.5281/zenodo.21297611 (DOI)