Tweet
Tweet

Slide 1

Slide 1 text

A new era has begun AI Assisted Software Dev Byron Saltysiak

Slide 2

Slide 2 text

About Me - Byron Saltysiak 25+ years in professional software development VP of Engineering at Authentica Solutions (EdTech) Now I enjoy building and talking about AI and LLM based apps https://www.linkedin.com/in/byronsalty/ Today’s slides and related links at: Byronsalty.com

Slide 3

Slide 3 text

Today’s Intuition

Slide 4

Slide 4 text

Today’s Intuition

Slide 5

Slide 5 text

Today’s Intuition

Slide 6

Slide 6 text

Today’s Intuition

Slide 7

Slide 7 text

Today’s Intuition

Slide 8

Slide 8 text

Today’s Agenda > Why Now Case Study Demo Tips & Lessons Learned Closing & QA

Slide 9

Slide 9 text

Why Now - Fast iteration Rise of Agents in Software Development: ● Cursor > Windsurf > VS Code & Zed ● MCPs (Model-Context-Protocol)

Slide 10

Slide 10 text

Why Now - Agentic IDEs are big

Slide 11

Slide 11 text

Why Now - LLM acceleration LLMs +++ ● GPT o1-mini ● Claude 3.5 ● Claude 3.7 ● Gemini 2.5 (In ~3 months)

Slide 12

Slide 12 text

Today’s Agenda Why Now > Case Study Demo Tips & Lessons Learned Closing & QA

Slide 13

Slide 13 text

Lactation Consultant Toolkit app

Slide 14

Slide 14 text

Lactation Consultant Toolkit app

Slide 15

Slide 15 text

Case Study: History Margaret Salty (my wife) identified a gap in the market. Initial Development: 1 weekend (November 2024) Initial Tools: VS Code + ChatGPT Tech stack: Flutter to target iOS and Android

Slide 16

Slide 16 text

Case Study: Today Initial Launch: December 2024 Dev Tools: Windsurf Tech stack: Flutter to target iOS and Android 800+ Apps Sold $4k+ Sales 17 countries

Slide 17

Slide 17 text

Today’s Agenda Why Now Case Study > Demo Tips & Lessons Learned Closing & QA

Slide 18

Slide 18 text

Demo

Slide 19

Slide 19 text

Today’s Agenda Why Now Tools Intro Demo > Tips & Lessons Learned Closing & QA

Slide 20

Slide 20 text

Tips - What AI does well (and doesn’t do well) 👍 Does Well 👎 Falls Short Speed Complex architecture Repetition Long-term context Boilerplate generation Nuanced trade-offs Suggesting alternatives Ownership & judgment

Slide 21

Slide 21 text

Tip 0: Use the IDE AI “Rules” Cursor rules engine is even better

Slide 22

Slide 22 text

Tip 1: Start with Documentation, not Prompts ● Don’t just ask AI to build a feature for you. ● Do ask AI to document the feature instead. Review the document and iterate, then ask it to implement. ● Do add documentation directly into your project with Markdown. Bonus Tip: Mermaid.js is a great addition to markdown documentation.

Slide 23

Slide 23 text

Tip 2: Use Git … more ● WIP commits, local branches, etc ● You’ll get overly ambitious and completely screw everything up. ● Get to a stable spot and commit so you have an easy rollback. Bonus Tip: LLMs are great at weird git scenarios

Slide 24

Slide 24 text

Tip 3: Encapsulation gives AI (and Humans) Focus ● Functional style ● Use standard libraries / components

Slide 25

Slide 25 text

Tip 3: Encapsulation - the Rules

Slide 26

Slide 26 text

Tip 4: Direct AI to use your test suite ● Automate verification of features ● You might need to tweak settings to allow some commands.

Slide 27

Slide 27 text

Tip 4: Direct AI to use your test suite

Slide 28

Slide 28 text

Bonus Tip: Lots more here… https://www.youtube.com/watch?v=CIAu6WeckQ0

Slide 29

Slide 29 text

Today’s Agenda Why Now Tools Intro Demo Tips & Lessons Learned > Closing & QA

Slide 30

Slide 30 text

Closing

Slide 31

Slide 31 text

Questions?

Slide 32

Slide 32 text

Appendix