Skip to content
DraconiticResearch
Search Sources API
203,952 studies31 sourcesUpdated 22 min ago
PreprintCC BY 4.0Zenodo

Autonomous Relationship Engine: Persistent Intent, Goal-State Evaluation, and Cognitive Memory for Enterprise Agentic Execution

  • Edison Ramón Sira Carreño
Open at sourcePDFDOICite

Abstract

Enterprise information systems remain strongly organized around persistent entities, CRUD operations, and prede-

fined workflow states. These mechanisms are effective for deterministic transactions, but provide limited architectural

support for interactions whose desired outcome is expressed as an evolving goal over heterogeneous context. Large

Language Model (LLM) agents provide adaptive planning and tool orchestration, yet their probabilistic execution

introduces challenges concerning bounded execution, authorization, state divergence, memory continuity, and au-

ditable control. This paper proposes the Autonomous Relationship Engine (ARE), a Design Science Research (DSR)

artifact centered on a persistent, time-bounded Intent Node as a unit of enterprise agentic execution. An Intent Node

is modeled as I(t) = <G, C, P, M(t), S(t)>, where Gt represents an operational goal specification, Ct hard constraints,

Pt runtime policy, Mt cognitive memory, and St authoritative execution-domain state. ARE distinguishes incoming

intent from an active commitment and defines commitment-scoped action admissibility over the active commitment,

authoritative state, action scope, and deterministic policy decision. Goal satisfaction is evaluated through &Phi;(Gt , St ),

while candidate actions are generated through probabilistic planning and subjected to a Policy Decision Point (PDP)

expressed through Policy-as-Code (PaC) before controlled execution. A tri-modal memory model provides episodic,

semantic, and procedural context while remaining non-authoritative for authorization. The paper formalizes the

commitment lifecycle, goal-state evaluation, bounded execution, policy enforcement, memory authority boundaries,

and a reproducible experimental protocol, including a falsifiable ablation of commitment-linked cognitive-context re-

construction after interruption, state change, policy update, and session resumption. The contribution is positioned

as an architectural synthesis and formalization rather than a claim of novel individual mechanisms or experimentally

established performance; the proposed evaluation criteria remain hypotheses for subsequent validation.

Related

  1. The moderating effect of cognitive re-evaluation on the relationship between superficial performance and job burnout in psychiatric nurses
    Shasha Hong, Shuhui Huang, Mianmian Zhu +4
    Frontiers in PsychiatryPeer-reviewedCC BY 4.0Full textPsychiatry & mental healthCrossrefEurope PMCPMC
  2. Architecting the Sustainable Enterprise: A Cognitive Load Framework for Managing Next-Generation Workforce Integration
    Mykhailo Zhylin, Viktoriia Zhylina, Kateryna Kryvobok +2
    ZenodoPreprintCC BY 4.0Full textZenodo
  3. Autonomous MTOR activation in COPD airway epithelium impairs resolution through persistent mucus cell metaplasia and lysosome dysfunction
    K. Kudrna, F. L. Vilches, X. Wang +4
    bioRxivPreprintAll rights reservedFull textGeneral medicinebioRxiv
  4. Investigating the dynamic relationship between anxiety and spatial memory using autonomous ecological momentary assessment
    C. Z. Han, K. C. Zhao, L. M. Wang +12
    bioRxivPreprintAll rights reservedFull textNeurosciencebioRxiv
  5. Dissociating certainty and uncertainty for passengers of autonomous vehicles: evidence against a perfect inverse relationship.
    Mervyn Franssen, Erik D van der Spek, Marieke H Martens
    Frontiers in psychologyPeer-reviewedCC BYFull textEurope PMC

Cite

BibTeXRIS
BibTeX
@misc{carreno2026autonomous,
  title = {{Autonomous Relationship Engine: Persistent Intent, Goal-State Evaluation, and Cognitive Memory for Enterprise Agentic Execution}},
  author = {Edison Ramón Sira Carreño},
  howpublished = {Zenodo},
  year = {2026},
  month = oct,
  doi = {10.5281/zenodo.23489471},
  url = {https://zenodo.org/records/23489471},
  note = {CC BY 4.0},
}
RIS
TY  - UNPB
TI  - Autonomous Relationship Engine: Persistent Intent, Goal-State Evaluation, and Cognitive Memory for Enterprise Agentic Execution
AU  - Edison Ramón Sira Carreño
PY  - 2026
DA  - 2026/10/20
PB  - Zenodo
DO  - 10.5281/zenodo.23489471
UR  - https://zenodo.org/records/23489471
AB  - Enterprise information systems remain strongly organized around persistent entities, CRUD operations, and prede-  fined workflow states. These mechanisms are effective for deterministic transactions, but provide limited architectural  support for interactions whose desired outcome is expressed as an evolving goal over heterogeneous context. Large  Language Model (LLM) agents provide adaptive planning and tool orchestration, yet their probabilistic execution  introduces challenges concerning bounded execution, authorization, state divergence, memory continuity, and au-  ditable control. This paper proposes the Autonomous Relationship Engine (ARE), a Design Science Research (DSR)  artifact centered on a persistent, time-bounded Intent Node as a unit of enterprise agentic execution. An Intent Node  is modeled as I(t) = <G, C, P, M(t), S(t)>, where Gt represents an operational goal specification, Ct hard constraints,  Pt runtime policy, Mt cognitive memory, and St authoritative execution-domain state. ARE distinguishes incoming  intent from an active commitment and defines commitment-scoped action admissibility over the active commitment,  authoritative state, action scope, and deterministic policy decision. Goal satisfaction is evaluated through &Phi;(Gt , St ),  while candidate actions are generated through probabilistic planning and subjected to a Policy Decision Point (PDP)  expressed through Policy-as-Code (PaC) before controlled execution. A tri-modal memory model provides episodic,  semantic, and procedural context while remaining non-authoritative for authorization. The paper formalizes the  commitment lifecycle, goal-state evaluation, bounded execution, policy enforcement, memory authority boundaries,  and a reproducible experimental protocol, including a falsifiable ablation of commitment-linked cognitive-context re-  construction after interruption, state change, policy update, and session resumption. The contribution is positioned  as an architectural synthesis and formalization rather than a claim of novel individual mechanisms or experimentally  established performance; the proposed evaluation criteria remain hypotheses for subsequent validation.
ER  -