Future of Coding with AI ft. Andrey Breslav
Summary
Andrei Breslav, creator of Kotlin, discusses spec-driven development with Codespeak, arguing that clear human intent is crucial as AI generates code. He explains how structured, AI-assisted specs replace vague prompts, allowing teams to review high-level requirements instead of massive code diffs. This approach helps catch inconsistencies early and ensures AI outputs align with actual business needs in complex software projects.
Topics discussed
Introduction: Andrei Breslav and the future of software development
Andrei's background: Kotlin, mental health, and CodeSpeak
Defining specs: From informal documents to formal languages
The spectrum of specifications and human intent
Example: Specifying Tic-Tac-Toe with evolving requirements
Semantic entropy and measuring requirement clarity
The bottleneck of code reviews and team context
Shifting focus from code to high-level requirements
AI-generated tests: Regression vs. acceptance testing
Abstraction levels: Assembly to high-level specs
Versioning specs and managing technical debt
Motivations for spec-driven development in teams
Programming language choice and token efficiency
The future workflow: Merging product and engineering roles
Listen ad-free on Castria