AI-Augmented Development

---

---

AI Summary

Webinar Summary: AI Augmented Software Development

Introduction

• Topic: AI augmented software development
• Focus: Developer role and AI’s impact on coding

Agenda

• Introduction and live demo
• Overview of generative AI in development tools
• Market and technology state
• Backend workings and adoption considerations

Presenter Background

• Not a full-time coder, background in research
• Proficient in C++ and Python, backend development
• AI tools are helpful for coding efficiency

Live Demo

• Demonstrated ChatGPT’s capabilities:
  • Generating data sets
  • Creating SQL queries
  • Generating Python code for plotting
  • Creating Java classes from XML
• Discussed dialogue-based interaction with AI
• Highlighted the need to validate AI suggestions

AI Tools Overview

• AI tools can make novel code suggestions
• Assist with tests and documentation
• Best with commonly used languages
• Require learning a few shortcuts

Drawbacks

• Suggestions are “hallucinated,” not guaranteed to be correct
• May suggest deprecated or suboptimal code
• Can propose solutions using bad practices
• Suggestions are based on existing code style

GitHub Copilot

• Offers code suggestions within IDE
• Can have a conversation about code
• Telemetry data sent back to GitHub for improvement

Market and Tools

• GitHub Copilot was the first integrated solution
• Other tools available, some allow self-hosting
• Open source alternatives are catching up

Backend Mechanics

• Large language models predict plausible next words
• Code suggestions are probabilistic, not deterministic
• Pre-processing and post-processing are necessary

SWOT Analysis

• Short-term strengths and weaknesses discussed
• Long-term opportunities include improved productivity and learning
• Threats include overreliance and potential legal issues

Smalls Research Initiatives

• Inventory of tools used
• Pilot with test licenses for a coding assistant
• Proof of concept for AI tools in other development lifecycle functions

Conclusion

• Coding without an assistant may soon feel outdated
• Correctness and productivity are key considerations
• Intellectual property and security are important concerns
• AI assistants will become part of default office software

Looking Ahead

• Co-piloting will be a theme for 2024
• Integration in Microsoft Teams and GitHub workspace
• Standardization and compatibility between tools are in development

Recommendations

• Avoid sensitive information in code accessible to AI
• Mark generated code and validate thoroughly
• Don’t accept code you don’t understand
• Maintain high standards to ensure quality suggestions

Contact

• For Smalls initiatives, email generative AI suggestions to generativeai@smalls.be

End of Summary