LN: Assalaarachchi et al. (2026) — Toward Agentic Software Project Management: A Vision and Roadmap

Bibliographic Reference

Citation: Assalaarachchi, N., Hoda, R., Hassan, A. E., & Grundy, J. (2026). Toward agentic software project management: A vision and roadmap. arXiv:2601.16392. ICSE 2026 AGENT Workshop. https://arxiv.org/abs/2601.16392


Pass 1 — Bird’s Eye View (5 Cs)

CAssessment
CategoryVision paper + roadmap
ContextFirst paper to explicitly propose “Agentic Project Manager” as a research direction
CorrectnessPosition paper; empirically-grounded vision based on current AI capabilities.
Contributions(1) Concept of “Agentic Project Manager” (APM) in software project management; (2) Taxonomy of autonomy levels (from assistant to junior PM agent); (3) Ethical implications of human role shift to “coach”; (4) Research roadmap for SPM 3.0
ClarityExcellent. Clear framework and diagrams.

Relevance: ⭐⭐⭐⭐⭐

This is the most directly aligned paper with PUMA’s long-term vision (Stage 5: Smart PMO). Pioneer paper in “agentic SPM” micro-niche.


Pass 2 — Key Points

The paper proposes Software Project Management 3.0: autonomous AI agents take over routine PM tasks while humans transition to strategic coaching roles. Three autonomy levels:

  1. Assistant: AI suggests, human decides (PUMA Stages 1–3)
  2. Co-pilot: AI leads, human validates (PUMA Stage 4)
  3. Junior PM Agent: AI executes autonomously, human coaches (PUMA Stage 5 vision)

Ethical framing: The human-as-coach model directly supports PUMA’s HITL principle and addresses concerns about job displacement.


Pass 3 — Virtual Reconstruction

Q1 (How can I use this?): Cite as the foundational vision paper for PUMA’s Stage 5 Smart PMO. The 3-level autonomy taxonomy provides the conceptual framework for PUMA’s incremental design (Strategies C → D → long-term).

Q2 (Does it do what it claims?): Vision paper, no empirical results. Consistent with current AI capabilities literature.

Q3 (What if?): What if the “junior PM agent” level requires a different evaluation methodology than the benchmark approach PUMA uses? A future study could evaluate APMs on full project simulations rather than individual task metrics.


PUMA Integration

MOCs