AI-Powered Software Development
This course covers the agent loop and the toolchains built on it, embeddings and attention, inference and reasoning, hallucination and evaluation, model comparison, tool contracts, and multimodal and realtime work — the mechanisms underneath building software with AI, each one rebuilt in code that runs in the page.
Prerequisites
- You ship in at least one language, and are comfortable with pull requests and CI
- Basic familiarity with HTTP APIs and relational databases; no machine-learning background assumed
- No API key required — the labs are cost models and simulations that run in the browser
Prerequisites
A production language, Git, code review & CI
Comfortable shipping in at least one language, and with pull requests and continuous integration
HTTP APIs & relational databases
Basic familiarity — no machine-learning background required
No API key required
The labs are cost models and simulations you run in the browser; nothing calls a live model
Lessons
AI Coding Agents Essentials
An agent is a loop — and the loop is where cost and failure live
Advanced AI Toolchains & Automation
Deterministic scaffolding around a non-deterministic worker
Transformers & Embeddings
Every layer of the stack was tuned on prose. Code is not prose.
Attention Mechanisms & Context
The n×n matrix is the thing everything else exists to avoid
Inference, Prompting & Reasoning
Three of the four things you would reach for produce confidence, not correctness
Hallucinations & Evaluation
The metric cannot see the failure it is trusted to catch
Comparing Models
The ranking you read was measured on someone else’s codebase
Function Calling & Tool Contracts
Every guarantee is about the form of the call, never its meaning
Multimodal & Real-Time
Each modality converts at its own rate and arrives on its own deadline
Unlocks
Steering an agent deliberately
Done conditions, tool schemas, and rules files stop being folklore once you can see the loop they act on
Judging an agent claim
Benchmark scores, context-window numbers, and cost comparisons all become checkable rather than persuasive