📊 The Catalyst: What Changed
⚡ The Shift: Before vs. After Opus 4.5
❌ Before Opus 4.5
- 💣 AI code came with "hidden bombs"
- ✍️ Significant manual coding still required
- 🔍 AI was a helper, couldn't carry the load
- 📝 Workflow: mostly writing code by hand
- 🐌 Slow onboarding for new learners
✅ After Opus 4.5 / 4.6
- 🏗️ Human work shifted to architecture
- 🤖 AI handles almost all manual coding
- 🧹 Human role: direct, verify, clean up
- 📖 Workflow: reading & reviewing code
- 🚀 Learners can be productive much faster
Reality Check: AI Still Can't Do Everything
In a real-world collaboration on a Stripe payment integration,
the AI couldn't piece together the complex, security-sensitive feature end-to-end.
It left behind: incomplete integrations, logic gaps, and leftover mess.
But here's the nuance: no manual code from scratch was needed.
What was needed: programming knowledge, syntax literacy, and architectural thinking.
The Lesson Learned
AI agent coding first, manual coding second.
Reading code first, writing code second.
The order of learning matters more than we think.
You can always go back and learn the fundamentals
once you're already productive.
💡 The Keyboard vs. Pen Analogy
Old Way: Pen First
Teaching learners to write every line by hand before using AI tools — like insisting on handwriting before allowing a keyboard.
New Way: Keyboard First
Start with AI tools to become productive fast. Visit "handwriting" later. Teaching the keyboard is faster and delivers more satisfaction.
🛤️ The New Learning Path
Learn AI Agent
Master AI-assisted coding first
Read Code
Understand & clean AI output
Debug & Verify
Build code literacy
Architecture
Tackle higher-order problems
Manual Coding
Learn the fundamentals deeper
🎯 The Three Pillars of the AI-First Pivot
AI Agent Coding First
Every learner knows how to use an AI agent for coding before they write a single line of code manually. The AI agent becomes their first tool, not an afterthought.
Start with Free AI Models
Learners use lower-end or free models like GitHub Copilot Free during training. The limitations of a lower-end model are actually a feature for learning.
Smart Assignment Strategy
Assignments pair learners with lower-end AI. The AI writes; the learner reads, cleans, and directs. Those who master a weak model become extraordinary with a strong one.
The Beautiful Payoff
A learner who can work effectively with a lower-end AI model
will be extraordinarily effective with a high-end model like Claude Opus 4.6.
They'll have the code literacy, the debugging instincts,
and the architectural awareness to leverage the best tools to their fullest potential.
Speed from AI + Depth from struggle = Unstoppable developer.
🌅 The Tết Takeaway
For Educators: The world has changed. AI isn't coming — it's here. The most responsible thing we can do is prepare learners for the world as it is, not the world as it was.
For Learners: Don't fear AI tools. Embrace them as your first instrument. The fundamentals don't disappear — they become more meaningful once you understand why they matter.
For High Achievers: Sometimes stepping away from the keyboard is exactly what you need to figure out the most important thing to type next.
The curriculum pivot begins now.
AI agent first. Reading first. Architecture first.
Manual coding? It can wait. 🚀