AI-Driven SDLC

Notes from the AI-Driven Software Development Life-Cycle course at SOICT, Hanoi University of Science and Technology, taught by Dr. Bùi Thị Mai Anh.

The through-line of the course is a question that tool demos tend to skip: AI can already accelerate almost every individual activity in the lifecycle, so why hasn't the lifecycle as a whole got faster? Each session works a running case study in teams rather than presenting conclusions.

Sessions

Session 1 — From Traditional SDLC to AI-Driven SDLC: Where Are We? — the context-drift problem, what AI does and what humans decide in each of the six phases, the HITL / HOTL / HOOTL models, and the two team workshops step by step.
Session 2 — AI-Driven Requirement Engineering and System Design — why AI that asks beats AI that answers, exploring three architectures instead of accepting one, context engineering, and the fourteen workshop steps against a running system.
Session 3 — AI-Driven Development and Quality Engineering — spending the spec: one coding task at a time, unit tests that check behaviour, a review where AI reports but does not fix, and the two quality gates that end in a human verdict.
Session 4 — AI-Engineering Governance — context debt, the failure taxonomy, the real case where the course's own sample solution was built from a superseded spec while its tests stayed 100% green, and six governance principles.
Session 5 — AI-Driven Workflow — the closing session: quality gate as sufficient evidence rather than passing tests, where AI's autonomy starts and stops, and five review rules for the AI era.

All thirty prompts in one place

Every prompt used across these sessions is collected in the Prompt Vault — full text, grouped by activity, each one linking back to the session it came from. There is a single-file download there too.

DevOps — the delivery tooling these processes run on
AI & Automation video summaries — agentic AI workflows in practice