What we believe
AI should accelerate careful work—not replace the habits that make work dependable. Clear intent, small changes, tests, and review remain the foundation.
Human Judgment. AI Speed. Better Software.
Code with AI
without losing the plot.
A practical guide to planning clearly, changing one thing at a time, testing what matters, and keeping yourself—not the model—in charge.
- 17
- practical questions
- 17
- deeper explanations
- 1
- human in charge
Stay in the loop
- 1 State the intent
- 2 Make a small change
- 3 Run the checks
- 4 Review the diff
The field guide
Questions worth asking
Choose a chapter or search for the problem in front of you. Each answer starts practical; the deep dive explains why it works.
About 10 minutes
The quick-start path
Seven questions that establish the safest and most useful working loop. Start here if the full guide feels like a lot.
Go to the sources
References and further reading
A curated starting point for the research, reporting, and practitioner perspectives behind this guide.
A living field guide
Practical, skeptical, and human-centered.
This project exists to help people use capable AI tools without handing over judgment, context, or accountability.
Who made it
Practical Vibe Coding brings together two complementary perspectives: applying AI responsibly in real organizations and understanding the computer science beneath the tools.
LinkedIn ↗
José De Jesús
José is CTO of AI & Automation for IBM Expert Labs, with 30+ years of enterprise architecture experience helping organizations move responsible, practical AI from possibility to production.
View José on LinkedIn →
LinkedIn ↗
Mark Wegman
Mark is an IBM Fellow and Chief Scientist for Software Technology at IBM Research, as well as an ACM and IEEE Fellow. His foundational work spans compilers, algorithms, universal hashing, and data compression.
View Mark on LinkedIn →Help it improve
Rate individual sections, leave focused suggestions, or send José and Mark general feedback about the overall site.
Acknowledgements
With thanks to Dave Pellecchia and everyone who has challenged an assumption, tested a workflow, or shared a better practice.