SkyWaysThe agentic manual

How AI-DLC, BMAD, Spec Kit and Scrum fit one lifecycle

Lay every method on the same four phases and two things appear: they mostly agree, and they all leave the same decisions open.

7 minutesIntermediateLesson 6 of 7Updated By

On this page
  1. Sound familiar?
  2. Where does each method sit?
  3. How to combine them, step by step
  4. Where you'll use it
  5. Why it matters
  6. Try it
  7. Key takeaways
  8. FAQ
  9. Apply it in your role
  10. Sources and credits
All lessonsStart here

Getting started

  1. What is the agentic PDLC?
  2. How this tutorial works

Agentic PDLC fundamentals

  1. The evolution of the PDLC
  2. Why agentic AI projects fail
  3. P0 · Frame
  4. P1 · Design & Spec
  5. The one hard gate
  6. P2 · Build & Prove
  7. P3 · Run & Learn
  8. The eight loops
  9. The evidence pack

Methods decoded

  1. AI-DLC vs AIDD vs agentic SDLC
  2. What is AI-DLC?
  3. What is AIDD?
  4. What is the BMAD Method?
  5. What is spec-driven development?
  6. One lifecycle for every method
  7. How much process a change needs

Running delivery

  1. How to run an agentic AI project
  2. Bolts vs sprints
  3. A board for agentic work
  4. Cut delivery from months to weeks
  5. Review AI code by risk
  6. How accurate must an agent be?
  7. Prove the agent meets its bar
  8. Shadow mode and cut-over
  9. Why the AI bill is 4× the estimate
  10. Guardrails that hold
  11. The five governance gates
  12. Catch AI drift
  13. AI incident postmortems

By role

  1. For product managers
  2. For program managers
  3. For solution architects
  4. For engineers
  5. For forward-deployed engineers
  6. For QA
  7. For DevOps and platform
  8. For business sponsors
  9. For executives

Teams and organisation

  1. Team structure for agentic AI
  2. Measuring AI productivity
  3. The maturity model
  4. Rolling it out in 90 days

Practice

  1. The SkyWays case study
  2. The operating rhythm
  3. The workbench
  4. Twelve exercises

Interviews and careers

  1. What is an FDE?
  2. Six answer frameworks
  3. AI PM interview questions
  4. FDE interview questions
  5. GenAI engineer questions
  6. Agentic engineer questions
  7. AWS GenAI questions

In short

Every building method fits onto the agentic PDLC's four phases, and most of their stages land in P1 and P2. Laying them side by side shows the same gap in all of them: none sets an acceptance bar per slice from what a mistake costs, and none enforces authority per action. Almost none says how to watch a model in production and report its value beside its cost.

Every methodon one lifecycleP0FrameAI-DLC · intentBMAD · analyst'sbriefShape Up · shaping,bettingP1Design & SpecAI-DLC · Mob ElaborationKiro · requirements, designSpec Kit · specify, planBMAD · PRD, architectureP2Build & ProveAI-DLC ·Construction, boltsKiro · tasksSpec Kit ·implementBMAD · stories, Build, reviewScrum · sprints become boltsP3Run & LearnAI-DLC · OperationsAIDDLC · Deploy & LearnThe methods mostly agree on the shape. What none of them settles is the bar, the authority and who watches theagent after launch.

Every method on one lifecycle

P0Frame
  • AI-DLC · intent
  • BMAD · analyst's brief
  • Shape Up · shaping, betting

P1Design & Spec
  • AI-DLC · Mob Elaboration
  • Kiro · requirements, design
  • Spec Kit · specify, plan
  • BMAD · PRD, architecture

P2Build & Prove
  • AI-DLC · Construction, bolts
  • Kiro · tasks
  • Spec Kit · implement
  • BMAD · stories, Build, review
  • Scrum · sprints become bolts

P3Run & Learn
  • AI-DLC · Operations
  • AIDDLC · Deploy & Learn

The methods mostly agree on the shape. What none of them settles is the bar, the authority and who watches the agent after launch.

In this lesson you'll learn:

  • where each well-known method's stages sit on the four phases;
  • the decisions every one of them leaves open, and who should own each;
  • how to combine a building method with the lifecycle without running two processes.

Sound familiar?#

  • The team runs Scrum, the architects want Spec Kit, and a vendor is pitching AI-DLC, and the argument is about which one wins.
  • Every method you evaluate has a planning phase and a building phase, and none has a clear answer to "is it good enough to ship?"
  • After launch, nobody's method says who watches the agent, or who tells the sponsor what it cost.

The map settles the first by showing that the methods mostly agree. It exposes the second and third as gaps that a lifecycle has to fill, whichever method you choose.

Where does each method sit?#

Four methodsonthe four phasesThe SkyWays PDLCP0 · FrameP1 · Design & SpecP2 · Build & ProveP3 · Run & LearnSDDestablishedThe spec, not the code, is what youmaintain; code is regenerated fromit.a spec sketch,lightly: the painand the ceilingthe spec is theartefact: eightfields, stableIDscode generatedfrom the spec,regenerated onchangelightly: theincident editsthe spec firstBMAD MethoddocumentedSkills and five named agents inyour coding tool; one Build loop,documents added as the workgrows.the Analyst'sbrief, or none atallPM, UX andArchitect: PRD,DESIGN.md, thespinea spec and itsstories, thenBuild per storyextended here:learn andadjust, into thenext briefAI-DLC (AWS)documentedThree phases and bolts of hours ordays replace sprints; a personapproves every boundary.inception: anintent becomesunits of workmobelaboration,NFRs capturedconstruction inbolts, mobconstructionoperations, runadaptivelyAIDD, the daily craftestablishedHow an engineer works with acoding agent day to day, whichevermethod frames it.silentsilentcontext files,story files, aharness in CI,review by risksilentThis manual addsworking methodThe parts none of the methodsdecide, in the phase where eachbelongs.the AI-fitverdict, theautonomyceiling, thevalue linewhat thesign-off needs:spec, bar,guardrails, andthe authoritybudgetone unknownper bolt, a lowerbound not ascore, theshadow runtwo numbers,drift, theincident as thenext P0sign-off (the hard gate)A filled cell is where the method says something about that phase; a dashed cell is where a team has to bring its own answer. The lastrow is where this manual's own devices sit.

Four methods on the four phases

establishedSDDThe spec, not the code, is what you maintain; code is regenerated from it.
  • P0 · Framea spec sketch, lightly: the pain and the ceiling
  • P1 · Design & Specthe spec is the artefact: eight fields, stable IDs
  • P2 · Build & Provecode generated from the spec, regenerated on change
  • P3 · Run & Learnlightly: the incident edits the spec first
documentedBMAD MethodSkills and five named agents in your coding tool; one Build loop, documents added as the work grows.
  • P0 · Framethe Analyst's brief, or none at all
  • P1 · Design & SpecPM, UX and Architect: PRD, DESIGN.md, the spine
  • P2 · Build & Provea spec and its stories, then Build per story
  • P3 · Run & Learnextended here: learn and adjust, into the next brief
documentedAI-DLC (AWS)Three phases and bolts of hours or days replace sprints; a person approves every boundary.
  • P0 · Frameinception: an intent becomes units of work
  • P1 · Design & Specmob elaboration, NFRs captured
  • P2 · Build & Proveconstruction in bolts, mob construction
  • P3 · Run & Learnoperations, run adaptively
establishedAIDD, the daily craftHow an engineer works with a coding agent day to day, whichever method frames it.
  • P2 · Build & Provecontext files, story files, a harness in CI, review by risk
  • Silent in P0, P1 and P3
working methodThis manual addsThe parts none of the methods decide, in the phase where each belongs.
  • P0 · Framethe AI-fit verdict, the autonomy ceiling, the value line
  • P1 · Design & Specwhat the sign-off needs: spec, bar, guardrails, and the authority budget
  • P2 · Build & Proveone unknown per bolt, a lower bound not a score, the shadow run
  • P3 · Run & Learntwo numbers, drift, the incident as the next P0

A filled cell is where the method says something about that phase; a dashed cell is where a team has to bring its own answer. The last row is where this manual's own devices sit.

Not competitors. Each method speaks to part of the lifecycle. The decision that matters is not which method but how deep to go on this change.
MethodP0 · FrameP1 · Design & SpecP2 · Build & ProveP3 · Run & Learn
AWS AI-DLCInception: the business intentInception: Mob Elaboration, units of workConstruction: Mob Construction, in boltsOperations: infrastructure and deployment
AIDDLCFoundation, Discovery & IntelligenceArchitecture & Design, SpecificationBuild, ValidationDeploy & Learn
Kirononerequirements.md, design.mdtasks.md, task by tasknone
GitHub Spec KitnoneConstitution, specify, plan, tasksImplementnone
BMAD MethodThe Analyst's briefThe PRD, the UX pair, the architecture spineA spec, its stories, Build per story, the retrospectiveExtended BMAD: learn and adjust, into the next brief
ScrumBacklog refinementSprint planningSprints (here, bolts) and the reviewThe retrospective
Shape UpShaping and the betting tableThe shaped pitchThe six-week cyclenone
Stage-gateDiscovery and scopingThe business caseDevelopment, testing and validationLaunch

How to combine them, step by step#

Step 1 · Map your method's stages onto the four phases#

Write your method's stages against P0 to P3, as in the table. Most land in P1 and P2, because most methods are about how to build. That is not a flaw. It is what they are for.

Step 2 · Find the empty cells and the thin ones#

Look for the phases your method leaves empty or covers only in part. For almost every building method the same things are missing, because they are questions about the model inside the product, not about how the product is built:

Open decisionPhaseWho should own it
Whether the job needs a model at all, and what it is worth net of running and checkingP0Product manager
The acceptance bar for each slice, from what a mistake costsP1Product manager and QA
Authority per action, with every limit enforced in a tool's signatureP1Solution architect
A harness that blocks a merge when a slice falls below its bar, and a shadow runP2Engineering and QA
A drift watch, and the saving reported beside the spendP3QA, DevOps and the sponsor

Step 3 · Give each gap an owner inside your existing ceremonies#

Do not add a second process. Put the bar and the authority budget into the planning ceremony you already hold; make the harness a required check in the pipeline you already run; add the drift chart and the two numbers to the review you already give the sponsor.

Step 4 · Keep one spec, whatever the method#

Every method above produces some form of specification. Keep one, in the repository, as the artefact agents build from, whether your tool calls it requirements.md, a spec, a PRD or a pitch. The eight fields of the agentic spec are what it must contain.

Step 5 · Hold the phase exits, not the method's ceremonies#

What makes it work is not the rituals but the exits: each phase ends on evidence, and P1 → P2 is the hard gate. A team can run Scrum, Spec Kit and AI-DLC's mob sessions and still skip every exit; a team that holds the exits can run any of them.

Where you'll use it#

  • When choosing a method: map the candidates first, and choose on what fits your team, since the gaps are the same.
  • When two teams run different methods: the four phases are the shared vocabulary for hand-offs.
  • When auditing a programme: an empty P3 column is the most common and most expensive gap.

Why it matters#

Method arguments are expensive and mostly beside the point. The failures that cancel agentic projects (escalating cost, unclear value, weak risk controls) live in the cells that every building method leaves empty. Mapping them makes that visible in one table.

ScrumSpec Kitwhich one wins?no methodwatches this
The argument is about which method wins. No method says who watches the agent once it is live.

Try it#

A team runs Scrum with Spec Kit inside its sprints. Its product is an agent that drafts and sends customer replies. Using the map, which phase has no owner at all?

Show the answer

P3, Run & Learn. Spec Kit stops at implement, and Scrum's retrospective looks at the team, not at the agent's behaviour in production. Nobody is watching the mix of the agent's replies for drift, nobody has rehearsed switching it back to drafting-only, and nobody reports what it saved beside what it costs. The fix is an owner for each inside existing ceremonies, not a new method.

Key takeaways#

  1. Every method fits the four phases, and most of their stages land in P1 and P2.
  2. They share the same gaps: the bar per slice, authority per action, proof before launch, and the operating loop.
  3. Combine without a second process: own the gaps inside your existing ceremonies, keep one spec, and hold the phase exits.

FAQ#

Can I use AI-DLC with Scrum?#

Yes. The ceremonies can stay; the planning unit shrinks from the sprint to the bolt, and Inception's Mob Elaboration can replace or feed sprint planning. The phase exits, especially the hard gate between design and build, apply whichever you keep.

What is the best AI development methodology?#

There is no best one in general. The methods mostly agree on how to build and differ in ceremony and tooling; choose the one that fits your team and tools. What decides success is whether the decisions they all leave open (the bar, the authority budget, the operating loop) have owners.

Is Shape Up compatible with AI-driven development?#

Yes. Shaping and the betting table are a strong P0; the shaped pitch is a P1 input; and the six-week cycle can be run as a sequence of bolts. Shape Up's appetite, how much time an idea deserves, pairs naturally with the value line.

What replaces the sprint when AI writes the code?#

The bolt: a slice of hours or days, carrying one unknown, integrated the same day. The term comes from AWS's AI-DLC; the rule of one unknown per bolt is this manual's.

Apply it in your role#

If you are…Do thisThe AI-augmented shortcut
A forward-deployed engineerMap the customer's method onto P0 to P3 in the first meeting, then fill the three gaps every method shares: the bar, authority in code, production evidence.Ask a model to place each stage of the customer's process on the four phases.
A product manager or FDPMRun mixed-method teams on one lifecycle: the phases and hand-offs are shared, the method inside each team is theirs.Have a model generate the hand-off checklist per phase for each method in use.
A GenAI or agentic AI engineerWhatever the method, the harness, the caps and the per-call log are the same code. Build them once, as a library.Ask a coding agent to extract them into a shared package with its own tests.

Across the enterprise. Let teams choose methods and standardise the hand-offs: four checklists, one evidence pack index and one portfolio report, whatever each team calls its stages.

The ten-minute workflow. One map of every method your organisation uses:

Our teams use <methods>. Map each method's stages onto P0 Frame, P1 Design & Spec, P2 Build & Prove
and P3 Run & Learn in one table. Then list what each method leaves unsaid on three points: an
acceptance bar per slice, authority enforced per action, and production evidence of value beside cost.

Sources and credits#

IdeaOriginSource
AI-DLC's phases, rituals and boltsBorrowedRaja SP (2025). AI-Driven Development Life Cycle. AWS
AIDDLC's seven phasesBorrowedAIDDLC: AI-Driven Development Lifecycle Standard
Kiro's specs and Spec Kit's workflowBorrowedKiro specs · GitHub (2025). Spec Kit
The BMAD MethodBorrowedBMad Code. BMAD-METHOD
Scrum's eventsBorrowedSchwaber, K. & Sutherland, J. (2020). The Scrum Guide
Shaping, betting and six-week cyclesBorrowedSinger, R. (2019). Shape Up. Basecamp
Stage-gateBorrowedCooper, R. G. (1990). Stage-gate systems. Business Horizons 33(3)
The mapping, and the gaps it showsOriginal: this tutorialThe Agentic PDLC

Prefer GitHub? This track is on the wiki, and this lesson is markdown. Found a mistake? Edit this lesson, or say so.