Everything, by category
Start
Your role, end to end
For leadership
Libraries
Play
Elsewhere
SkyWaysThe agentic manual
SimulatorPlay

The tutorial

Agentic PDLC Tutorial: Run AI Agent Projects, Step by Step

Short lessons on running software projects where an AI model does part of the work: what to decide, in what order, and how to know you got it right.

Start with lesson one Start from your role

55 lessons8 tracksabout 7 hours in allfree

Who this page is for, and how to use it
For

Anyone running, building, testing, operating or funding software where a model does part of the work. and anyone preparing for an interview for such a job.

Use it to

Learn the agentic PDLC in five-to-ten-minute lessons, in order, then apply each one in your own role with the shortcut and the ten-minute workflow at the end of every lesson.

How
  1. New? Read Getting started, then Fundamentals, top to bottom.
  2. In a hurry? Use Start where you are below to jump to your role's lesson.
  3. Every lesson ends with Apply it in your role and a prompt to paste; every diagram is drawn live.

Eight tracks, in order

  1. 1Getting startedWhat the agentic PDLC is, in one sitting, and how to get the most out of the lessons that follow.2 lessons · about 17 min · starts with “What is the agentic PDLC?”
  2. 2Agentic PDLC fundamentalsThe 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 lessons · about 67 min · starts with “The evolution of the PDLC”
  3. 3Methods decodedAWS 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 lessons · about 49 min · starts with “AI-DLC vs AIDD vs agentic SDLC”
  4. 4Running deliveryThe 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 lessons · about 85 min · starts with “How to run an agentic AI project”
  5. 5By roleHow 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 lessons · about 58 min · starts with “For product managers”
  6. 6Teams and organisationHow to structure teams, measure productivity without fooling yourself, assess maturity by controls rather than tools, and roll the method out in ninety days.4 lessons · about 28 min · starts with “Team structure for agentic AI”
  7. 7PracticeThe SkyWays case study in thirteen episodes, the operating rhythm from daily to quarterly, the workbench, and twelve exercises with worked answers.4 lessons · about 30 min · starts with “The SkyWays case study”
  8. 8Interviews and careersWhat 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 lessons · about 73 min · starts with “What is an FDE?”

In short

This free tutorial teaches the agentic PDLC, a four-phase lifecycle (P0 Frame, P1 Design & Spec, P2 Build & Prove, P3 Run & Learn) for products in which a model drafts, decides or acts, in five-to-ten-minute lessons. It works with whatever method you already use, including AWS AI-DLC, the BMAD Method, spec-driven development and Scrum. The home page and the method page call it the SkyWays PDLC; the two names mean the same lifecycle.

Each role's part in the four phases is drawn on the method page.

Start where you are#

If you are…Start withBecause
New to agentic deliveryWhat is the agentic PDLC?Everything else assumes the four phases
A product managerFor product managersThe pain, the autonomy and the bar are yours
A programme or delivery managerFor programme managersThe board, the cadence and the review queue
A solution architectFor solution architectsWhich steps may guess, and where the caps live
An engineer or engineering leadFor software engineersBolts, the harness and the merge gate
A forward-deployed engineerThe forward-deployed engineer guideFrame, Deliver, Evolve: twelve steps, each with its template
In QAFor QA"It works" becomes proof that it works
In DevOps or platformFor DevOps and platformThe gateway, the flags and the rollback
A sponsorFor business sponsorsTwo numbers, on one line, every cycle
An executiveFor executivesFour decisions nobody else can make
Leading the changeRolling it out in 90 daysOne feature, five phases, one trap each
Wanting practiceThe SkyWays case studyNinety days, thirteen episodes, failures left in
Preparing for an interviewSix answer frameworksThen the bank for your role, with strong answers
Curious about forward-deployed workWhat is an FDE?The role, its three stages, and who works beside it
Wondering why these projects go wrongWhy agentic AI projects failThe failure modes are new, and quiet
About to write a specThe eight-field specThe step comes with its template
About to launchShadow, then five percentThe order a cut-over goes in
Asked for a business caseThe value lineThe arithmetic is worked
Running a workshop37 scenariosTwenty sectors to choose from

What you will be able to do#

By the end of the fundamentals you will be able to:

  1. Frame an AI feature as a measured pain, decide whether it needs a model at all, and set how much the machine may do on its own.
  2. Specify it so a coding agent, or an engineer who was not in the room, can build it without asking, with an acceptance bar per slice.
  3. Build and prove it in bolts, against a bar, with evidence a sceptic accepts.
  4. Run it: cut over safely, catch drift, and report what it saved beside what it cost.

The role, organisation and practice tracks then apply the same four phases to your job, to how a whole organisation adopts them, and to a full ninety-day case with twelve exercises; the interview track turns all of it into answers for AI product, FDE, GenAI, agentic and AWS roles.

How the lessons work#

Every lesson has the same slots in the same order: the answer first, a picture, the problem stated plainly, the idea step by step, a problem to try, three takeaways, and its sources, each marked Borrowed, Adapted or Original to this manual. Why it is built that way.

The lessons are written once and published twice: here, with the live diagrams, and on the GitHub wiki, where they sit beside the manual's full reference pages.

Also on the wiki, and as llms.txt for AI assistants.