Multi-criteria IT project resource assignment method by a collective of specialized ai agents

Abstract

The paper proposes a multi-criteria IT project resource assignment method performed by a collective of specialized artificial intelligence agents. The method is intended for local or partial replanning after an event violates a previous assignment, deadline, workload or risk restriction. Instead of selecting a candidate by one centrally calculated score, the approach separates the assessment of competence, direct cost, workload and accumulated fatigue, execution risk, and coordination. Each agent produces an explainable proposal with constraints; the coordinator resolves compatible preferences through bounded negotiation and rechecks feasibility after every concession. A resource is represented by a dynamic profile containing competences, rate, current workload, accumulated fatigue, execution risk and availability. First, the feasibility agent creates a candidate set that satisfies mandatory skill, capacity and fatigue requirements. Then the specialized agents issue comparable local scores and confidence values. The influence of an agent depends on current confidence, historical predictive performance and data freshness, which reduces the impact of obsolete project information. The result includes a selected assignment, reserve candidates, active constraints, collective confidence and an escalation route for a project manager. A simulation study used sixty base project graphs, six disruption types and thirty controlled initial states for each scenario. The method was compared with unchanged assignment, maximum-competence selection, minimum-cost selection, workload balancing and a centralized weighted criterion. Relative to the strongest baseline, the centralized criterion, the proposed method decreased normalized actual loss by eleven point sixty-four percent and hard-constraint violations by thirty-six point eighty-four percent. Ablation confirmed the separate contribution of competence, fatigue, risk and negotiation components. The method can therefore serve as an explainable resource-assignment module in an adaptive IT project management loop.