TDD & Generative AI: A Perfect Pairing
summary
Claim: tests become the sole spec while an LLM writes implementations and the process is automatically verified. Built a CLI jar plus a Maven plugin that calls local Llama 3.1 (and Cloud Anthropic/OpenAI), runs up to 5 code loops × 5 test loops = 25 attempts, compiles and JVM-loads generated classes, and executes JUnit 5 tests. Demos include odd/even, prime lists and Spring Boot endpoints; simple cases succeed quickly, framework and project‑context cases reveal limits and need more tests or stronger models. Automated TDD→generation proof‑of‑concept
AI-written summary · about
chapters (23)
- 0:00Introduction to the talk
- 1:53Reverse engineering tests with AI
- 2:58Explaining the three concepts
- 5:25Introducing test driven generation
- 8:41Why verifying AI output matters
- 9:25Happy flow demonstration
- 11:35Simple demo with DeepSeek
- 13:44Prime number generator demo
- 15:30Automating manual steps
- 16:55Building the test driven generator
- 21:54Working with local LLMs
- 23:22Tool demo with prime numbers
- 25:08Adding features via tests
- 27:39Real code with Spring Boot
- 31:31Limitations and Maven plugin
- 33:17Spring Boot endpoint demo
- 35:26Age calculation endpoint
- 38:19Cloud LLM upgrade demo
- 40:28What works and what doesn't
- 41:51Cost and efficiency analysis
- 43:16Future improvements
- 44:41Vision for IDE integration
- 45:48Conclusion and resources
related talks
46mTDD & generative AI - a perfect pairing?
48mTDD & generative AI - a perfect pairing?
45mTDD & generative AI - a perfect pairing?
46mGenerative AI: Say Goodbye to Deterministic Testing?
35mTDD with AI?
01h 06mDoes GenAI Mean the Death of Test-Driven Development (TDD)
21mLLM-Powered Unit Testing: Writing Reliable Java Tests with AI Assistance
40mDavid Fogl: Leveraging Generative AI for Enhanced Software Testing