What is AI-DLC? The AI-driven development life cycle, explained
AI-DLC is AWS's method for building software with an AI agent in the lead and people approving each step. This guide covers what it is, how its phases and gates work, and what it leaves for the people doing the approving.
The short answer
AI-DLC stands for AI-Driven Development Life Cycle. It is a software delivery method that AWS introduced in July 2025. An AI agent drives the work: it plans, asks questions, proposes designs, and writes code and tests. People stay in charge of the decisions, and the work does not move forward until a person approves it. You will also see it written as AIDLC or AI DLC.
AI-DLC is not the same thing as “AI SDLC”. AI SDLC is a general phrase for any software development lifecycle in which AI does part of the work. AI-DLC is one specific, named method, with its own phases, vocabulary and open-source tooling.
Where it came from
Raja SP, a Principal Solutions Architect at AWS, set out the method on the AWS DevOps blog on 31 July 2025. His argument is that the two common ways of using AI in development both fall short. In AI-assisted development, AI helps with single tasks such as code completion or documentation, and the process around it stays the same. In AI-autonomous development, AI generates whole applications with no person involved. AI-DLC sits between them: AI leads the workflow, and people make the decisions that need business context.
The post describes one repeating pattern. The AI creates a plan, asks clarifying questions, and implements only after a person has validated the plan. The post presents a method and reports no measured results. Its claim on speed is that work that took weeks can be done in hours or days.
The vocabulary
AI-DLC renames several familiar things. These four terms come from the AWS post.
- Mob Elaboration. The whole team reviews the AI's questions and proposals during Inception.
- Mob Construction. The team gives real-time input on technical and architectural decisions during Construction.
- Bolt. Replaces the sprint. A shorter cycle, measured in hours or days.
- Unit of Work. Replaces the epic.
The phases
The method names three phases.
- Inception. The AI turns business intent into requirements, stories and units of work, and the team validates them.
- Construction. The AI proposes architecture, domain models, code and tests, and the team decides the technical questions.
- Operations. The AI manages infrastructure as code and deployments, with the team overseeing.
AWS Labs also publishes open-source workflows, under the MIT-0 license, that run the method inside a coding agent. Version 2.11.1, current when we read it on 19 November 2026, describes five phases and 33 stages. It has setup instructions for Claude Code, Kiro, Codex CLI, Cursor, opencode and GitHub Copilot.
The five phases in the AI-DLC workflows
| Phase | Stages | What happens |
|---|---|---|
| Initialization | 3 | The workspace is set up and its state file is created. This phase runs without approval gates. |
| Ideation | 7 | The intent is captured and framed, with market research, feasibility, scope and rough mockups. |
| Inception | 9 | Requirements analysis, user stories, domain design, units of work and delivery planning. On a project with existing code, the code is reverse-engineered first. |
| Construction | 7 | Functional and non-functional design, then code generation for each unit, then build and test and a CI pipeline. |
| Operation | 7 | Deployment, environments, observability, incident response and performance validation. Every stage in this phase is conditional. |
Where the checks are
The workflows have two kinds of gate.
Approval gates. Every stage outside Initialization ends with one. The person approves or requests changes, and the workflow needs a real message from that person before it continues.
Verification gates. These sit at three phase boundaries: Ideation to Inception, Inception to Construction, and Construction to Operation. They are automated. They check that the artifacts the finished phase should have produced exist, that the links between them hold (each requirement maps to a story, for example), that nothing is orphaned, and that related artifacts are consistent with each other.
Requirements get particular attention. In the Requirements Analysis stage the agent writes its questions to a file with lettered answer options, and it does not proceed on partial answers. It then reads the answers for vague language, contradictions and missing detail, and asks follow-up questions where it finds any.
Later, the Build and Test stage runs once every unit has been coded, and produces test results and a quality report.
AI-DLC puts a person at every gate. That is its strength, and it is also where the work lands.
What the method asks of the approver
Three questions are worth asking before you adopt it.
1. What does the approver check a requirement against?
The verification gate at the end of Inception confirms that every story traces to a requirement. The approval gate asks a person whether the requirements are right. That person is reading a document an AI drafted in minutes, often under time pressure. A score for each requirement on defined measures, with the open points listed, gives the approver something to approve against.
2. Where does a business decision live after the work is done?
The workflows keep a decision log, decisions.md, for design decisions and their rationale. Business decisions are a different kind of record: what “fast” means for a search page, which customers a rule applies to, who decided and when. In the workflow they are made in answer to the agent's questions. Ask where those answers will be when the next requirement in the same project raises the same question.
3. Who signs that the running software does what was decided?
Build and Test reports on code the AI wrote, using tests produced in the same workflow. For a release that a business owner, an auditor or a control function has to accept, ask what evidence they will be shown, and whether they can read it.
Where Spec2TestAI fits
Spec2TestAI is our product. It is independent of AWS and of AI-DLC, and it works at the same three points.
- Before a requirement is approved. Requirements Intelligence checks every story on 32 quality measures. What it leaves open becomes a question for a person who can decide.
- After a question is answered. The answer is kept in the project knowledge base as a decision, with who decided and when, and informs later requirements in the same project. That record is Specification Intelligence.
- Before release. Tests and synthetic test data are generated from the requirement, and Spec2RunAI runs them in plain English against the live application. Every run ends in a Run Evidence Report.
Spec2Code AI delivers the approved requirement, your standards and your security policy into GitHub Copilot, Cursor, Windsurf, Cline and Claude over MCP, so the agent builds from what was approved.
How these fit together across the lifecycle is on Quality for the AI SDLC.
What this does not tell you
AI-DLC is one vendor's method, not an industry standard. Treat the name as a useful label for AI-led delivery with human approval, and judge the practice on its results in your own teams.
AWS's post gives no measured comparison with other methods. We have not run a controlled comparison of AI-DLC with and without requirement scoring, and we are not aware of a published one.
The workflows change quickly. This article describes version 2.11.1 as we read it on 19 November 2026.
Sources
- SP, R., “AI-Driven Development Life Cycle: Reimagining Software Engineering”, AWS DevOps & Developer Productivity Blog, 31 July 2025. Definition; the two approaches it contrasts; the three phases; Mob Elaboration, Mob Construction, bolts and units of work.
- AWS Labs, aidlc-workflows repository, README read 19 November 2026. Version 2.11.1; MIT-0 license; five phases and 33 stages; the coding tools it supports.
- AWS Labs, AI-DLC workflows guide, “Phases and Stages”, read 19 November 2026. Stage lists; approval gates; verification gates at three phase boundaries; Build and Test.
- AWS Labs, AI-DLC workflows stage reference, “Inception”, read 19 November 2026. The Requirements Analysis questions file and follow-up questions; the Inception phase check;
decisions.md.
Get your backlog scored
Send one backlog. We score it on 32 measures, find the conflicts across the stories, and list the questions nobody has answered yet. Fixed fee.
See the Quality Risk Audit