Changing behavior with AADV (AI Assisted Development with Verification) new post on Substack!. Adding new behaviors means regenerating entire modules. How do you begin to accomplish that in a well-designed, multi-module solution?.

Examples: The lingua franca of AADV? . New AI Code Correct substack post!. When doing AI-assisted development with CAX, examples are the basis for compiling intent into code and tests, like a compiler produces executable artifacts.

"Behavioral drift in AADV". LLMs hallucinate intent, too, during code generation. Why it's a problem and some thoughts on addressing it.

AADV: Better vibes than (bad) vibe coding .

Streamlining the CAX cycle

Many of the skills needed to succeed at AI-assisted development (with verification!) aren't terribly daunting. Learning how to supply examples to the LLM is at the core of AADV; this AI Code Correct post provides an introduction.

Not "all vibe coding is bad," but defining a dev process in a tweet is irresponsible. We know how it will mostly turn out. AI Code Correct post: "Squelching bad vibe coding," in which I review my approach to AI-assisted development.

AI Code Correct: Avoid bad vibe coding!. Latest post: A (free) follow-up to the CAX-driving of a cyclomatic complexity module. When CAXing a cyclomatic complexity module, why did the LLM generate a non-compliant cyclomatic solution, failing two tests?.

RT @pragprog: Spring Sale Week Two.Text, Tools, and Testing.40% Off – This Week Only!.Code: 2025testing at checkout. Titles on sale this we….

AI Code Correct: "CAXing a cyclomatic complexity model.". The LLM generates a solution that fails errant examples. Fixing them gets me to 100% compliance. Next up: Thoughts on why this occurred.

Promo code: 2025TESTING.

Week two of the Pragmatic Bookshelf's Spring Sale starts Wednesday April 16! Pick up a copy of Pragmatic Unit Testing in Java (3e) at a 40% discount! .

Do well-designed LLM solutions comply more? ..(Proposed research for LLM code generation via AADV/CAX. A true teaser--no answer yet. Help shape the research!).

"Prompt, Pray, Repeat"--latest AI Code Correct post , in which I make a hypothesis that does not hold true. The LLM does respond really well to a prompt including some specifics about the solution's design, though.

"Prompt, Pray, Repeat"--my latest AI Code Correct Substack post , in which I make a hypothesis that does not hold true. The LLM does respond really well to a prompt including some specifics about the solution's design, though.

Your LLM-generated code *will* inevitably demonstrate a compliance gap. At least you'll have tests that fail, if you're practicing CAX. What then, though?. In this post: a simple example gone wrong, and a thankfully near-effortless resolution.

Looking for a new exercise for interviewing, a practice problem to learn a new language, or a first intro to TDD? The name normalizer kata:. Solutions in many languages, Python conspicuously missing. I'd love a good contribution.

Create, Assess, eXecute (CAX)

we must start breaking the loop: Help feed the AI models with better code, via AAD.

RT @bookauthority: Great work @jlangr! We're honored to inform you that "Modern C++ Programming with Test-Driven Development" made it to Bo….