The Augmented Human: Taming the Beast, the Art of Staying in Control
L'Humain Augmenté : Dompter la bête, l'art de rester maître
A pragmatic look at AI in development: how to harness it without losing control. Built around 3 golden rules and real before/after examples from 15 months of hands-on AI usage.
“AI proposes, the Augmented Human decides.”
The 3 Golden Rules
The 90/10 is real
AI gets you 90% in 10 seconds. Your 10% takes 90% of your attention. Knowing where you are in that ratio is the critical skill.
Never trust the re-generated
Screenshot, lock it down, supply the stable parts yourself. Once AI regenerates something you considered final, it's fair game for drift.
Triangulate before you validate
3 experts are better than 1. Truth lives at the intersection. Never ship on a single model's opinion.
Presentation Highlights
Before/After: AI Design Evolution
The "IA ou pas IA?" game — a 15-month progression of AI-generated design, from garbled text to near-perfect output, shown live with audience participation.
The FunnySlide
A Claude Sonnet model confidently telling me that Claude Sonnet 4.6 doesn't exist — while running as Claude Sonnet 4.6. Concrete proof that triangulation isn't optional.
Key Concepts
The 90/10 Rule
AI gets you to 90% in seconds — the critical last 10% is where your expertise and judgment make or break the outcome.
Design by Algorithm — Screenshot & Replace
A workflow for using AI to generate UI/content, then locking the good parts with screenshots before regenerating context around it. Prevents unintended drift.
Le Conseil des Sages — Multi-Model Triangulation
Generate with one model → Critique with a second → Validate with a third → YOU synthesize. The disagreement between models is where the real insight lives.
Architecture Levels
Code correctness → Architecture resilience → "What will BREAK and WHEN?" — each level requires a different type of AI prompt and a different level of human oversight.
Chaos Engineering with AI
Using AI to proactively find failure modes before production does. Ask models to attack your own architecture, not just validate it.
The Boeing 737 MAX Analogy
Understanding your tools is not optional. Automation without comprehension is how systems fail catastrophically. Be the brake, not just the accelerator.
Related Content
Building Andon-SSP with AI
The 4-part blog series that the presentation draws from — deep dives into the same principles with real code examples.
Read the blog series →“AI proposes, the Augmented Human decides.”