Tech Excellence

As a developer community, we don't agree about the viability of LLMs to produce production-worthy code. To make things worse, some headstrong developers are charging forward with little more than (what they hope to be good) vibes. Yet we know that LLMs hallucinate, and we know that they produce low-quality code. We TDD practitioners also know that it's a bad idea to trust our ability to gate defects solely via reading or manually testing code.

In his AI Code Correct substack, Jeff Langr has been currently exploring the realities and limitations of LLM-generated code through his AADV process.

In this session, you'll see perhaps dicey demonstrations of generating code to help you learn about AADV--a simple, example-driven approach. You'll be able to try it out yourself the moment you leave the session.

Outline of the session:

- What you get from an LLM when exuding (bad) vibes
- The prime directive of AI-generated code
- What’s that LLM really doing?
- The Create-Assess-eXecute (CAX) cycle
- Providing examples & vetting tests
- Improving quality with a style
- ZOMBIES
- The compliance gap: Dealing with defective code
- Design still matters. You still matter.

This presentation will be a mix of slides and demonstration (a risky proposition given the non-deterministic behavior of an LLM!).

ABOUT JEFF

Jeff Langr, who has been building software professionally since 1982, is the author of the AI Code Correct Substack, in which he explores how to produce deliverable AI-generated code.

He’s also a co-author of Bob Martin’s best-selling book, Clean Code, plus five books of his own on software development, including Agile Java: Crafting Code With Test-Driven Development, Modern C++ Programming With TDD, Agile in a Flash (with Tim Ottinger), and most recently, the 3rd edition of Pragmatic Unit Testing in Java. At his site, you’ll find hundreds of posts and links to 125+ articles published elsewhere.

A 2nd edition of Clean Code (with completely rewritten chapters from Jeff) coming soon!