In short
AI-DLC is AWS's 2025 methodology in which AI proposes and people decide across Inception, Construction and Operations, and AIDDLC is a separate seven-phase standard with a near-identical name. AIDD is the everyday craft of building with AI tools, agentic SDLC is a vendor term for agents across the software lifecycle, and spec-driven development and BMAD are ways to structure that building. The agentic PDLC is a product lifecycle for software that has a model inside it, which is the one question the others leave open.
Six terms two questions
- AWS AI-DLCAI proposes, people decide
- AIDDLCa seven-phase standard
- AIDDthe everyday craft
- Agentic SDLCagents in every phase
- Spec-driven devthe spec comes first
- BMAD Methodfive agents, one Build loop
and then
- The agentic PDLChow right, who may act, when it stops being true
The six on top answer how to build with AI. The one below answers what changes when the product itself decides. You need both.
In this lesson you'll learn:
- who coined each term, and what exactly it names;
- the one question that sorts them: is AI building the software, or is AI inside it?
- which to reach for when, and how they combine rather than compete.
Sound familiar?#
- A vendor deck says "AIDLC", a colleague says "AI-DLC", and a job posting asks for "AIDD", and nobody is sure they mean the same thing.
- You adopted a spec-driven tool, and it still had no opinion on how accurate the agent had to be.
- Every framework you read explains how to build faster; none says how to tell whether the thing you built is right.
The names are fragmenting faster than the practice. Sorting them by the question each one answers makes the choice simple.
What does each term mean?#
| Term | Coined or used by | What it names | Answers the question |
|---|---|---|---|
| AI-DLC | AWS: Raja SP, July 2025 | A methodology: AI plans and proposes, people validate and decide, across Inception, Construction and Operations, in bolts of hours or days | How do we build software with AI doing most of the work? |
| AIDDLC | aiddlc.ai, a separate standard | A seven-phase lifecycle: Foundation, Discovery & Intelligence, Architecture & Design, Specification, Build, Validation, Deploy & Learn (as its site listed them on 3 October 2026) | The same question, with more phases and gates |
| AIDLC | Various vendors and writers | Usually a spelling of AI-DLC; sometimes "AI development lifecycle" in general | Depends on the author: check which one they mean |
| AIDD | Widely used; no single author | AI-driven development: the everyday craft of building with AI tools (context files, story files, coding agents, review) | How do I work with coding agents day to day? |
| Agentic SDLC | Tooling vendors | The software lifecycle with agents taking part in every phase, from planning to operations | Where in delivery can agents act? |
| Spec-driven development | Kiro and GitHub Spec Kit, 2025 | Write the spec before the code, and keep it as the artefact agents build from | What should an agent build from? |
| BMAD Method | BMad Code | Skills in a coding tool, five named agents and one Build loop; every decision a document | How do we give AI building structure and an audit trail? |
| Agentic PDLC | This manual | A four-phase lifecycle, P0 to P3, for products that have a model inside them | How right must it be, who may authorise each action, and how will we know when it stops being true? |
The one question that sorts them#
Step 1 · Ask whether AI is building the software or is inside it#
Almost every term above is about AI as the builder: a coding agent writes, tests and deploys code that, once shipped, behaves like any other software. Those methods speed up building.
The agentic PDLC is about AI as part of the product: the shipped software itself calls a model to rank, draft, decide or act. That software is right only a share of the time, fails without an error, and changes behaviour when the world moves. Building it faster does not answer whether it is right.
Step 2 · Notice that most real projects are both#
A team building a rebooking assistant uses coding agents to write it (AI builds it) and ships a product that calls a model to rank options and issue refunds (AI is inside it). So the methods are not rivals. AI-DLC or spec-driven development can run the building inside P1 and P2, and the agentic PDLC supplies what they leave open: the acceptance bar per slice, the authority budget, the shadow run, the drift watch and the two-number report.
Four methods on the four phases
- 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
- 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
- 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
- P2 · Build & Provecontext files, story files, a harness in CI, review by risk
- Silent in P0, P1 and P3
- 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.
Step 3 · Pick by the question you are stuck on#
- Stuck on how to structure work with coding agents → spec-driven development, AI-DLC or BMAD.
- Stuck on daily habits (why the agent keeps ignoring conventions) → AIDD's context and story files.
- Stuck on whether it is good enough to launch, and who decides → the agentic PDLC.
Where you'll use it#
- When two people use the same acronym differently. Ask which question their method answers.
- When evaluating a tool. A spec-driven IDE decides where the spec lives, not what its bar is.
- When writing a job description or a training plan, so "AIDD" and "AI-DLC" name what you mean.
Why it matters#
A team that thinks it has adopted a lifecycle when it has adopted a building method will ship fast and discover the missing half in production: no bar, no authority budget, no drift watch. Keeping the two questions separate is how you notice which half you have.
Try it#
A team says: "We use Kiro specs and an AI-DLC workflow, so we've got our agentic lifecycle covered." Their product is a support agent that issues account credits. What have they probably not decided yet?
Show the answer
Everything about the model inside the product. Kiro and AI-DLC structure how the software is built: requirements, design, tasks, bolts. Neither decides how accurate the credit decisions must be for each kind of case, which credits need a person to approve, where the credit cap is enforced, how the agent will be run in shadow before it acts, or how drift in its decisions will be caught. Those are the agentic PDLC's questions, and they sit alongside the building method rather than replacing it.
Key takeaways#
- AI-DLC is AWS's 2025 method; AIDDLC is a different standard; AIDD is the everyday craft. Check which one people mean.
- Most terms describe AI building the software; the agentic PDLC describes software with AI inside it.
- Real projects are both, so the methods combine: a building method inside P1 and P2, the lifecycle around it.
FAQ#
What is the difference between AI-DLC and AIDLC?#
Usually none: "AIDLC" is most often a spelling of AWS's AI-DLC, the AI-Driven Development Life Cycle introduced in 2025. Some authors use AIDLC for "AI development lifecycle" in general, so check the source. AIDDLC, with a second D, is a separate seven-phase standard.
Is AIDD the same as AI-DLC?#
No. AIDD, AI-driven development, is a broad term for building software with AI tools day to day. AI-DLC is a specific methodology from AWS with named phases, rituals and bolts. You can practise AIDD without adopting AI-DLC, and AI-DLC relies on AIDD habits.
What is the difference between the SDLC and the PDLC?#
The software development lifecycle runs from requirements to deployment and maintenance of software. The product development lifecycle is wider: it begins with whether to build at all and ends with what the product earned against what it cost. The agentic PDLC is a PDLC because its first phase asks whether a job needs a model and its last reports value beside cost.
What about the agentic STLC?#
The STLC is the software testing lifecycle, so the agentic STLC, which some teams write as A-STLC, is testing with agents in it and testing of software that has a model in it. It is a slice of the product lifecycle, and in this manual it is the QA lead's eight steps: the proof each kind of step owes and the golden set in P1, the harness wired in as P1 turns into P2, the shadow run at the end of P2, the drift watch in P3.
Which should I use: AI-DLC, spec-driven development or the agentic PDLC?#
Every team that builds with coding agents needs spec-driven development and AIDD, every day. Add BMAD when the work crosses teams or an auditor reads it, and AI-DLC when you cannot tell how deep a change goes before you start. Put the agentic PDLC around all of it whenever the shipped software calls a model.
Apply it in your role#
| If you are… | Do this | The AI-augmented shortcut |
|---|---|---|
| A forward-deployed engineer | When a customer says "we do AI-DLC", ask which one: AWS's methodology, the AIDDLC standard, or everyday AI coding. Each needs a different first week. | Ask a model to map the customer's own vocabulary onto these terms and list where they collide. |
| A product manager or FDPM | Settle each team's building methods by the rule above, and keep one lifecycle for the portfolio. Filling the gaps every method shares is worth more than comparing methods. | Have a model draft a one-page glossary for your organisation, with each term's source and owner. |
| A GenAI or agentic AI engineer | Treat every method as a build discipline. The bar per slice and authority in code are yours to add under any of them. | Ask a coding agent to add a bar file and cap tests to whichever method's templates you use. |
Across the enterprise. Publish one terms page so that "AI-DLC" means one thing inside the organisation. Name confusion costs weeks in procurement, planning and vendor evaluation.
The ten-minute workflow. Untangle the vocabulary in a document or a vendor pitch:
Our teams use these terms: <list, each with an example sentence>. For each, say which it most likely
means (AWS AI-DLC, the AIDDLC standard, AIDD, agentic SDLC, spec-driven development or the BMAD
Method) and where the usage is ambiguous. Draft a one-paragraph definition we could adopt for each.Sources and credits#
| Idea | Origin | Source |
|---|---|---|
| AI-DLC: Inception, Construction, Operations; bolts; AI proposes, people decide | Borrowed | Raja SP (2025). AI-Driven Development Life Cycle. AWS |
| AI-DLC described as a development methodology coined at AWS | Borrowed | IBM. What is AI-DLC? |
| AIDDLC's seven phases | Borrowed | AIDDLC: AI-Driven Development Lifecycle Standard |
| Spec-driven development and its three levels | Borrowed | Böckeler, B. (2025). Understanding spec-driven development |
| The BMAD Method | Borrowed | BMad Code. BMAD-METHOD |
| Agents across every phase of the SDLC | Borrowed | CodeRabbit. A guide to the agentic SDLC |
| Sorting methods by "AI builds it" versus "AI is inside it" | Original: this tutorial | The Agentic PDLC |