AI Product Engineering
Building Intelligent Applications with Modern AI
Block 1: Foundations & The Exocortex | 2 Hours
Building AI required expensive GPUs, specialised teams, and years of training.
That world no longer exists for most practitioners.
Foundation models are accessible via a simple API call.
| Level | Example | Result |
|---|---|---|
| Consumer | "Give me a recipe" | One answer, conversation ends |
| Prompt Writer | "Act as a chef and…" | Better answer, still one-shot |
| Engineering Partner | Iterative: define → generate → evaluate → revise → test | Working software |
This course operates at Level 3.
True or false — cold-call, then tap.
True or false — cold-call, then tap.
True or false — cold-call, then tap.
🧠 THINK — Explore problems with AI
📚 KNOW — Build your knowledge base
🔨 BUILD — Scaffold with AI coding partner
🔌 CONNECT — Wire to cloud AI APIs
🛠️ EQUIP — Add tools & agentic workflows
🧪 EVALUATE — Test for failures
🚀 SHIP — Deliver working prototype
Classify each activity into the course framework stage.
| Term | Meaning |
|---|---|
| Foundation model | Pre-trained general capability |
| API | Structured request/response to the model |
| Token | Unit of text (cost + context) |
| Inference | Running the model for a response |
| Context window | How much text the model sees at once |
node -v.js, type function get — Copilot suggestsOne project for all 15 sessions. Answer all four:
| Question | Strong answer sounds like… |
|---|---|
| Problem | Specific waste of time / failure mode |
| User | Named cohort (not "anyone") |
| AI Role | What the model actually does |
| Success | Measurable (e.g. 8/10 grounded answers) |
Checkpoint: 60-second pair pitch — partner names one weakness.
Next Session: Prompting & Structured Outputs