Expert Panel Rewriting Software with AI Agents, GenAI & What’s Next? | WSO2Con Barcelona 2025



AI Summary

Panel Discussion on AI and Multi-Agent Systems

Panelists Introduced:

  • Malit Jing: VP of Research at WSO2, focusing on AI for the past 5 years.
  • Victor Dia: Principal Research Software Engineer at Microsoft Research, working on the Autogen framework for building AI agents.
  • Yasaman Kazeni: Director of Data Science at Capital One, with experience in AI solutions across retail and finance.

Key Discussion Points:

  1. Introduction to AI Agents:
    • Transitioning from traditional programming to multi-agent systems.
    • Importance of programming-level abstractions for building agent systems.
  2. Development and Production Challenges:
    • Challenges of moving AI models from research to production highlighted by Yasaman Kazeni.
    • Need for alignment between the AI solution and business needs, including user experience considerations.
  3. Evaluation Metrics for AI Systems:
    • Importance of robust evaluation metrics during development and after deployment.
    • Iterative improvements based on user feedback and telemetry data.
  4. Production Challenges Highlighted:
    • Examples of predictive models that required extensive integration into business processes (e.g., personalizing user experiences in retail).
  5. Multi-Agent System Design:
    • Recommendations for using pipelines and workflows for reliability.
    • Potential future shift towards more autonomous AI systems.
  6. Product Development Insights:
    • Corporate-level AI teams ensuring best practices and cross-compatibility across business units.

Conclusions:

  • Focus on user needs and problems to be solved, beyond just implementing AI technology.
  • Need to ensure that AI solutions fit well within existing workflows, emphasizing user experience, and effective metric evaluation to assess real-world performance and improvements.