Recoleta
GitHub
Source note

Now's the Time: Computer Science Must Evolve to Emphasize Software and Systems Engineering with Artificial Intelligence (AI)

Authors
Chandra N. Sekharan, George K. Thiruvathukal
Source
arXiv
Published
Collection
Software Intelligence
Cs EducationAI Assisted CodingSoftware EngineeringSystems EngineeringAgent Orchestration
Open arXivMarkdown source

Summary

This position paper argues that CS degrees should teach programming, data structures, and algorithms as foundations for building and checking AI-enabled systems, rather than as the main professional endpoint.

Problem

  • AI code assistants now handle many routine coding tasks, so graduates need skills in system design, verification, deployment, cost control, security, and human oversight.
  • Current CS curricula often train students to implement isolated code and canonical data structures, while employers need engineers who can ship, scale, secure, and maintain systems.
  • LLM-based systems can change behavior when models are updated even when APIs stay stable, which makes semantic regression testing and operational monitoring important.

Approach

  • The paper proposes reframing core CS topics as design primitives used inside larger software and AI systems.
  • It calls for more curriculum time on architecture, agent orchestration, domain integration, reliability, observability, cloud cost, security, ethics, and production delivery.
  • It recommends capstones with real users, cloud budgets, CI/CD pipelines, monitoring, architecture documents, security reviews, and post-mortems.
  • It treats LLMs as swappable but behaviorally unstable components, so students should learn behavioral regression testing, shadow deployment, and explicit checks above the API layer.

Results

  • The paper reports no new experiments, datasets, benchmarks, or measured model results.
  • It cites Gartner’s claim that within a few years most enterprise software engineers will use AI code assistants daily, but the excerpt gives no exact percentage.
  • It cites McKinsey & Company’s 2025 work claiming gains of tens of percentage points in productivity, time to market, customer experience, and software quality when AI is embedded across the product-development lifecycle.
  • It cites Stanford Digital Economy Lab work reporting a relative employment decline for early-career workers aged 22–25 in AI-exposed jobs, including software development, with reduced hiring as the main adjustment.
  • Its strongest concrete claim is curricular: CS programs should make AI-enabled systems engineering, verification, deployment, and ownership central requirements rather than electives.

Link