This dissertation presents the design, implementation and evaluation of a lightweight
Agentic Artificial Intelligence (AI) prototype intended to automate and support key aspects
of task management within software project workflows. The proposed system leverages the
capabilities of large language models (LLMs) to read and interpret software project
backlogs, perform task assignments, recommend deadlines, generate sprint-level
summaries, and prioritize work items.
The implemented prototype operates as a prompt-engineered, AI-assisted reasoning pipeline
within predefined boundaries. It demonstrates limited agentic characteristics, such as
contextual interpretation, goal-oriented task analysis, and structured recommendation
generation, but it should not be understood as a fully autonomous AI agent. Instead of
completely replacing human action, the system is designed to assist managers and team
leaders in planning. The prototype integrates agentic behaviors such as contextual
reasoning, goal awareness, and adaptive prioritization, allowing it to respond dynamically
to different project environments and backlog characteristics.
To ensure ethical compliance, the system was evaluated using synthetically constructed
backlogs inspired by real-world software projects, without relying on proprietary or
sensitive corporate data. The evaluation focuses on functional accuracy, responsiveness,
adaptability, and consistency of outputs across multiple project scenarios.
Beyond technical implementation, we examine the managerial implications of deploying
agentic AI systems in engineering management contexts. Key considerations include their
potential to enhance decision quality, optimize resource allocation, and improve team
coordination, alongside challenges related to trust, transparency, accountability, and
effective human–AI collaboration. The findings contribute to both the technical feasibility
and the strategic governance discussion surrounding the integration of autonomous AI
agents into modern software project management practices