Learn the agentic PDLC

A free tutorial in short lessons on running software projects where an AI model does part of the work: the four phases, the one hard gate and the eight loops; how AWS AI-DLC, AIDD, the BMAD Method and spec-driven development fit onto them; what changes for each role, from engineer to executive; how an organisation adopts it; a ninety-day case with twelve exercises; and interview question banks with strong answers for AI product managers, forward deployed engineers, GenAI, agentic and AWS roles. Each lesson opens with the answer, carries a picture, ends with how to apply it in your role, and credits its sources.

▶ Start here · or go straight to What Is the Agentic PDLC? The P0 to P3 Framework Explained

Track	What it covers	Lessons
Getting started	What the agentic PDLC is, in one sitting, and how to get the most out of the lessons that follow.	2
Agentic PDLC fundamentals	The four phases, the one hard gate, the eight loops and the evidence that crosses each hand-off: the model every other track builds on.	9
Methods decoded	AWS AI-DLC, AIDD, the BMAD Method and spec-driven development: what each one is, who coined it, where it sits on the four phases, and what it leaves for you to decide.	7
Running delivery	The practical core: the twelve steps, bolts and boards, cutting delivery time, reviewing AI-written code, setting and proving the bar, launching safely, cost, guardrails, governance, drift and postmortems.	13
By role	How each discipline works in the agentic PDLC (product, programme, architecture, engineering, forward-deployed, QA, DevOps, the sponsor and the executive): what changes, what is theirs, and what is not.	9
Teams and organisation	How to structure teams, measure productivity without fooling yourself, assess maturity by controls rather than tools, and roll the method out in ninety days.	4
Practice	The SkyWays case study in thirteen episodes, the operating rhythm from daily to quarterly, the workbench, and twelve exercises with worked answers.	4
Interviews and careers	What a forward deployed engineer does, six frameworks for answering AI interview questions, and deep question banks with strong answers for AI product managers, forward deployed engineers, GenAI engineers, agentic AI engineers and AWS roles.	7

55 lessons so far. The lessons live on the site, with live diagrams: akash-coded.github.io/…/learn. The wiki keeps this index, the reference pages, the course companion and the cohort kit.

---

The map

Five surfaces, ~800 files and a wiki that is now a manual in its own right. This page exists because no single surface can point at the other four.

Surface	What it is	Best entry point
📚 Curriculum	16 modules · 102 notebooks · 92 exercises · 63 solutions	START-HERE
🧭 Field guide	77 reference pages — frameworks, runbooks, playbooks, interview guides	Frameworks index
🧪 L.A.B. Simulator	Auto-graded labs · Learn → Apply → Break	python labs/runner/labctl.py next
💬 Discussions	66 threads, all tagged by track and level	Index of every exercise and lab
🛫 SkyWays, the agentic manual	The agentic PDLC as a manual, a game and a workbench: thirteen episodes, eight loops, nine simulations, seventeen calculators	Open it · pitch an idea
🧭 The role journeys	Five roles walked end to end, each with templates and copy-paste prompts at every step	PM · SA · Eng · QA · Ops
👤 The roles, defined	What each role is accountable for, may settle alone, and must not touch — plus how the role fails	PM · SA · Eng · QA · Ops · Sponsor
📘 The manual on the wiki	The same method in writing: four phases, eight loops, five delivery roles and the sponsor, nine how-tos, 31 exercises	The Agentic PDLC

If you would rather look than read

Four drawings carry the whole method, and each one links back into the page that explains it.

The picture	What it answers
The agentic PDLC	Four phases, one hard gate (the sign-off before anything is built), and the line that comes back
Eight loops make the line a ring	Which loops close themselves, and which three nobody is waiting on
Your role, across the four phases	One row per role, one column per phase — including the cells that should be empty
Where the model helps, and where it must not	The model drafts, you check, and the one thing per step that is never delegated

New here and want one link? START-HERE.

---

Pick a way in

I want to…	Go
Understand the concepts	Curriculum — read the module README first, always
Prove I can build it	L.A.B. Simulator — graded, offline, no AWS account
Look something up while working	Field guide
Fix an error right now	Error Index
Fit this around a job	Study Plans
Know what it will cost	Cost Log
Find a specific thing	Where do I find…?
Walk one feature through the whole lifecycle, by role	SkyWays, the agentic manual — in the browser, nothing to install
Walk my own role, step by step, with templates	The journeys — pick your role
Learn the method as a written reference	The Agentic PDLC → The Eight Loops → your role page
Work out a number — a bar, a bound, a bill	Formulas & Calculators
Settle a recurring argument	Decision Trees
Build intuition rather than follow a procedure	Mental Models — twelve shapes
Explain all of this to a board	The operating protocol
Run a workshop or an interview	Scenario Library · Exercises & Answers
Ten minutes, one idea, graded	A drill — post /drill AGL-101 in the Arena
See who is doing hands-on	Scoreboard · Hands-on Tracker
Get a lab graded without installing anything	Simulator Arena — post a comment, a bot replies
Ask a person	Q&A — check the answered ones first

---

The three things worth knowing before you start

1. Model access is not automatic. It is granted per model, per region, on request, and approval is not always instant. Start AWS setup before you need it.

2. Some of it bills for existing, not for use. A classic OpenSearch Serverless collection holds a minimum capacity, an AgentCore Runtime instance bills from boot until it is stopped, and stored long-term memory bills by the hour. They are what people leave running by accident. Teardown checklist.

3. A lot of this needs no AWS account at all. Modules 00, 01 and 15, every L.A.B. lab, rag_by_hand.py and quality_gate.py. That is 20+ hours of real work at zero cost while your model access is pending.

---

What is on this wiki, and what is not

The repository is canonical: versioned, reviewed, and it ships with the code. The site is where the tutorial and the manual live, indexed and searchable. This wiki is the layer between them, in five sections that the sidebar follows:

Section	What it holds	Why it lives here
Start	Start Here and one short page per tutorial track, each linking its lessons on the site	A reader who lives on GitHub still gets every lesson in order
Method and reference	The method pages, the journeys, the roles, nine how-tos, formulas, trees, scenarios, exercises	The method in writing. It changes as the practice does, and anyone can correct a number
Course and labs	The course companion: a page per module with its reading order, and the labs companion	Generated from the repository, with the tutorial lesson and workbench tool beside each module
Cohort kit	Eight ninety-minute sessions and a template, for anyone running this with a team	Teaching material is edited by the people teaching it
Community and maintenance	Error Index, Community Answers, Field Notes, Cost Log, region notes, the Roadmap, the runbook	Grows with no PR; wrong for a week is acceptable

Pages marked generated at the top are written by a script from a source in the repository; edit the source, not the page, and the Wiki edits workflow will tell you where if you forget. Everything else is hand-written and yours to improve.

Anyone can edit these. That is the point. How to.