Software Design: Humans vs AI
summary
A live classroom game called "Software Ball" runs on phones to measure code cost by having players pick simple versus generic snippets and combine them to meet user stories. The talk compares small local LLMs (7B on ~8GB GPU) and larger models (Gemini, GPT Mini/5, Coin 3) on generating refactorings, finding many fail to apply design principles in abstract scenarios (Claude notably never finishes), while human teams win a rematch decisively (best AI score ~82), concluding that using design principles rather than premature generalization is key.
AI-written summary · about
chapters (12)
- 0:00Humans vs AI
- 2:57Good software design
- 3:19Game rules with Alice and Bob
- 7:19Robot ball passing
- 12:24User story complexity
- 26:03Results: humans vs AI
- 30:23Design principles: SRP & OCP
- 33:39Decoupling and DI
- 36:48AI performance surprises
- 39:14Focus on principles
- 40:17Personal background
- 41:40Q&A
related talks
15mUsing AI effectively in large Java codebases
48mMastering the AI Shift: Adaptive Strategies for Software Evolution
45mThe AI Revolution in Software Development: From Code Writing to Quality Assurance
30mBuilding the systems that build the software
25mThe role of software design in an AI world
56mFundamentals of Software Engineering In the age of AI
01h 28mFuture of Software Programming with AI Bots at Hand
34mArtisinal Intelligence