From Junior to Senior: Allocating Agency and Navigating Professional Growth in Agentic AI-Mediated Software Engineering

Paper Title

From Junior to Senior: Allocating Agency and Navigating Professional Growth in Agentic AI-Mediated Software Engineering

Publication Info

• Topic area: The impact of agentic AI on software engineering practices, professional growth, and mentorship across experience levels.
• Keywords: Agentic AI, generative AI, software engineering, mentorship, professional growth, agency allocation, junior engineers, senior engineers, AI-mediated workflows, prompt reviews.

Background and Problem

• Problem / challenge: The integration of agentic AI into software engineering disrupts traditional workflows, career growth, and mentorship dynamics. Juniors face challenges in balancing learning with productivity, while seniors must adapt their roles to guide AI-native juniors effectively.
• Significance: Understanding how AI reshapes agency, skill development, and mentorship is critical to sustaining the talent pipeline and ensuring engineers retain control and accountability in increasingly AI-mediated environments.
• Motivation and related work: Prior research has explored generative AI in software engineering, but gaps remain in understanding how agency is distributed between humans and AI, particularly across experience levels. This study builds on prior work by investigating how juniors and seniors interact with AI tools, navigate professional growth, and maintain agency in their workflows.

Solution

• Proposed approach: A three-phase mixed-methods study combining interviews, task-based activities, and artifact reviews to analyze how junior and senior engineers allocate agency, perceive professional growth, and approach mentorship in AI-mediated workflows.
• Novelty:
  1. Identification of organizational policies as preconfiguring agency boundaries before individual preferences.
  2. Insights into divergent strategies for agency allocation between juniors and seniors in familiar and unfamiliar tasks.
  3. Proposal of Prompt & Code Reviews (PCRs) to preserve agency and accountability in AI-mediated workflows.
  4. Exploration of how AI impacts professional growth and mentorship dynamics across experience levels.
• Procedure and key techniques:
  • Phase 1: Interviews with senior engineers using ACTA and a Delphi-inspired process to design a realistic debugging task.
  • Phase 2: Junior engineers completed the debugging task with Cursor AI, followed by surveys and interviews.
  • Phase 3: Senior engineers reviewed junior artifacts (code, prompts, and reflections) to evaluate AI’s role in mentorship and learning.

Results

• Concrete findings:
  • Seniors maintain control through detailed delegation and iterative refinement, while juniors oscillate between over-reliance and cautious avoidance in unfamiliar contexts.
  • AI tools accelerate learning but cannot replace foundational understanding or system-level thinking.
  • Juniors report imposter syndrome and fragile understanding when relying heavily on AI, while seniors emphasize critical thinking and judgment.
• Advantage over baselines:
  • The study highlights nuanced differences in agency allocation and professional growth strategies between juniors and seniors, which were not addressed in prior research.
  • Proposed PCRs offer a structured approach to maintaining accountability and agency in AI-mediated workflows.
• Experiments / evaluation:
  • Phase 1: Five senior engineers identified tacit knowledge and debugging strategies using ACTA and Delphi methods.
  • Phase 2: Ten junior engineers completed a debugging task with Cursor AI, revealing diverse AI interaction patterns and confidence levels.
  • Phase 3: Five senior engineers reviewed junior artifacts, providing mentorship insights and feedback on AI usage.
• Limitations and future work:
  • Small sample size and regional concentration (USA and Canada) limit generalizability.
  • Differences in tasks between juniors and seniors prevent direct comparisons.
  • Future research could explore longitudinal studies, larger samples, and in-situ organizational settings to validate findings.

Summary

This study investigates how agentic AI reshapes software engineering practices, focusing on agency allocation, professional growth, and mentorship across experience levels. It reveals that organizational policies preconfigure agency boundaries, seniors leverage foundational instincts to maintain control, and juniors face challenges in balancing learning with productivity. The proposed Prompt & Code Reviews (PCRs) aim to preserve accountability and agency in AI-mediated workflows. While AI accelerates learning, it cannot replace the tacit knowledge transfer essential for system thinking and architectural judgment. These findings highlight the need for evolving mentorship practices and organizational support to sustain a robust talent pipeline in an AI-driven industry.