SingaDEV - Kevin Aubry - AI coding assistants

Slide 1

Slide 1 text

Scaling AI Coding Assistants SingaDev – Feb 2025

Slide 2

Slide 2 text

Software engineer from Switzerland 10 years in Singapore CTO of Palo IT Singapore Over 15 years in the industry Led multiple large-scale tech projects Guiding organizations on AI Coding Assistant adoption Passionate about DJing, dancing, and snowboarding ABOUT ME

Slide 3

Slide 3 text

We’re an international consultancy using tech as a force for good. We specialize in Human-centered design, agile software development, and transforming forward- thinking companies. INTRODUCTION TO PALO IT We believe that technology should have a positive impact on the world

Slide 4

Slide 4 text

No content

Slide 5

Slide 5 text

CONTENTS Why AI Coding Assistants Matter Real Productivity Gain vs Marketing Hype How to Scale AI Coding Assistants Effectively Tips & Tricks for Daily Efficiency Future Outlook Q&A 01 02 03 04 05 06

Slide 6

Slide 6 text

DIFFERENT CODING ASSISTANT

Slide 7

Slide 7 text

DIFFERENT CODING ASSISTANT GH Copilot Windsurf Cursor Trae Tabnine

Slide 8

Slide 8 text

WHY AI CODING ASSISTANTS MATTER Reduce Cognitive Load Free developers to focus on higher-level design rather than low-level code. Enforce Consistent Standards Maintain uniform coding styles and best practices across the team. Relief from Repetitive Work Automate tasks like documentation, test scaffolding, and boilerplate. Empower Developers Boost efficiency and excitement— developers can’t imagine going back. Enhance Code Understanding Quickly grasp and improve legacy or unfamiliar codebases.

Slide 9

Slide 9 text

REAL PRODUCTIVITY GAIN VS MARKETING HYPE

Slide 10

Slide 10 text

Slide 11

Slide 11 text

FROM A REAL PROJECT – FORECAST 0 20 40 60 80 100 120 week 0 week 1 week 2 week 3 week 4 week 5 week 6 week 7 week 8 week 9 week 10 week 11 week 12 Initial Scope Traditional Delivery Progress These data are derived from a real project’s data, but have been modified to protect confidentiality.

Slide 12

Slide 12 text

FROM A REAL PROJECT – ACTUAL 0 20 40 60 80 100 120 week 0 week 1 week 2 week 3 week 4 week 5 week 6 week 7 week 8 week 9 week 10 week 11 week 12 Delivered Scope, % Initial Scope Traditional Delivery Progress Actual Scope These data are derived from a real project’s data, but have been modified to protect confidentiality.

Slide 13

Slide 13 text

SUCCESS STORY 4 engineers 2 engineers + AI

Slide 14

Slide 14 text

OUR JOURNEY TO SCALE AI ASSISTANTS - BACKGROUND ~ 420 engineers globally ~ 120 engineers in Singapore AI training - internal & external - since 2023 Multiple project delivered with AI in the center Running transformation with clients > 1000 devs

Slide 15

Slide 15 text

OUR JOURNEY TO SCALE AI ASSISTANTS Phase 3 Phase 2 Phase 1 License Distribution Gave everyone access without guidance Low Adoption, Negative Feedback Training Sessions Basic introduction to features Slight Improvement; teams still hesitant Coaching & Experimentation On-the-ground support with real tasks; encouraged trying new prompts, exploring features Adoption boost, positive feedback

Slide 16

Slide 16 text

LESSON LEARNED – DO’S Offer Ongoing Coaching Frequent touchpoints and support foster true adoption. Train in Real Projects Use authentic tasks, not just theory or sample exercises. Encourage Experimentation Allow freedom to tweak prompts and explore new use cases. Identify Champions Empower early adopters to guide others and spread best practices. Track & Celebrate Wins Share success stories to maintain motivation and momentum.

Slide 17

Slide 17 text

LESSON LEARNED – DON’TS Assume Immediate Value Simply handing out licenses rarely leads to meaningful results. Rely Solely on Short Sessions One-off trainings raise awareness but don’t shift daily habits. Underestimate the Need for Context Developers can’t fully leverage AI without real-world examples and demos.

Slide 18

Slide 18 text

TIPS & TRICKS – DAILY EFFICIENCY Give Enough Context For AI to understands the bigger picture Refine Your Prompts Be explicit about what you need Validate & Review AI is non-deterministic and can hallucinate IDE Integrations Seamless experience enhance the workflow

Slide 19

Slide 19 text

TIPS & TRICKS – MINDSET & COLLABORATION Team Sharing Share successful prompts and reinforce consistent best practices Stay Curious Experiment with new features and updates Be Positive Use it as an accelerator, not a crutch

Slide 20

Slide 20 text

FUTURE OUTLOOK Evolving AI Capabilities • Expect deeper and wider context awareness • Automated debugging, QA, design, monitoring & operations. Shifting Bottlenecks • With coding sped up, processes like Product Ownership, Design, and QA become the new bottlenecks. • Future AI tools will help streamline these areas as well. Changing Skill Sets • Developers role will evolve and focus less (if any) on coding • AI fluency becomes a must-have skill rather than a “nice-to- have.” Competitive Advantage • Early adopters maintain a lead as AI tech continues to improve. • Teams that hesitate risk falling behind as AI becomes the new industry norm.