What is AI-DLC? AWS's AI-driven development lifecycle

In this lesson you'll learn:

• AI-DLC's three phases, its two mob rituals, and what a bolt and a unit of work are;
• how its open-source adaptive workflows decide how much process a task gets;
• where it sits on the agentic PDLC, and the decisions it leaves to you.

Sound familiar?#

• The coding agent builds a story in hours, then waits two weeks for the sprint review.
• Requirements are "clarified" in a thread nobody else reads, and the agent builds its own interpretation.
• Every task, from a typo to a new service, gets the same ceremony, so the team starts skipping it.

AI-DLC was written for exactly these three: planning cycles too long for how fast AI builds, clarification that happens in private, and process that does not scale with the task.

What is AI-DLC?#

AI-DLC, the AI-Driven Development Life Cycle, is a software engineering methodology introduced by Raja SP of AWS in July 2025. Its central inversion is who starts the work: the AI initiates, producing plans, questions and implementations, and people validate, making the decisions that need judgement, accountability or context the AI does not have. It describes itself as a machine-human collaboration, where earlier methods such as Scrum were designed around people doing all of the work.

How AI-DLC works, step by step#

Step 1 · Inception: turn intent into units of work, together#

The team states a business intent, and the AI turns it into requirements, stories and units of work: AI-DLC's replacement for epics. It does this through Mob Elaboration: the whole team in one session, answering the AI's clarifying questions and validating its proposals in real time, instead of each person reviewing a document alone.

Step 2 · Construction: AI proposes the design and the code, the team decides#

Using the validated context from Inception, the AI proposes a logical architecture, domain models, code and tests. Mob Construction is the equivalent ritual: the team clarifies technical decisions and architectural choices as they arise. Engineering disciplines such as domain-driven design and test-driven development are part of the workflow rather than optional extras.

Step 3 · Operations: deploy from the context you built up#

In Operations the AI applies everything accumulated in the earlier phases to infrastructure as code and deployment, with the team overseeing. The context is the thread that runs through all three phases: each one hands the next a record rather than a conversation.

Step 4 · Work in bolts, not sprints#

A bolt is AI-DLC's unit of iteration: a shorter, more intense cycle measured in hours or days rather than weeks. When an agent can build a story in an afternoon, a two-week sprint leaves it idle for most of the fortnight; the bolt matches the planning cycle to the building speed.

Step 5 · Let the workflow size itself#

In November 2025 AWS open-sourced AI-DLC's adaptive workflows, which choose both the breadth of a task (which stages to include) and the depth of each stage, from the complexity of the intent. A simple defect fix skips elaborate requirements analysis; a new service gets the full treatment. The project, awslabs/aidlc-workflows, is MIT-0 licensed and runs inside several coding agents, including Claude Code, Kiro, Cursor, Codex and GitHub Copilot.

How AI-DLC fits the agentic PDLC#

AI-DLC is a method for building with AI. The agentic PDLC is a lifecycle for products with AI inside them. They combine rather than compete:

The reason both are needed is that AI-DLC tells you how to build fast and deliberately, and says little about how right a product that calls a model must be, or who may authorise each of its actions. If your shipped software only runs deterministic code that an agent wrote, AI-DLC may be most of what you need. If it calls a model, you need the rest.

Where you'll use it#

• When your team builds with coding agents and the sprint no longer matches the building speed.
• When clarification keeps happening in private threads: mob sessions make it shared and recorded.
• When process is one-size-fits-all: adaptive depth is the fix, whichever method you run.

Why it matters#

AI-DLC is the most complete published answer, from a major vendor, to how software teams should reorganise when AI does most of the building. Knowing precisely what it covers, and what it does not, lets you adopt it without assuming it also decides your acceptance bars and guardrails.

Key takeaways#

1. AI-DLC is AWS's 2025 methodology: AI initiates and proposes, people validate and decide.
2. It runs Inception, Construction and Operations, with mob rituals, units of work and bolts of hours or days.
3. It is a building method; for software with a model inside it, pair it with a product lifecycle such as the agentic PDLC.

FAQ#

Is AI-DLC only for AWS?#

No. It is a methodology, not a service. Its open-source adaptive workflows run inside several coding agents, including Claude Code, Kiro, Cursor, Codex and GitHub Copilot, and nothing in it requires AWS infrastructure.

What is Mob Elaboration?#

The Inception ritual in which the whole team works through the AI's clarifying questions and proposals together, in real time, as the AI turns a business intent into requirements, stories and units of work. Mob Construction is the equivalent during Construction, for technical decisions.

What is a bolt in AI-DLC?#

A bolt is AI-DLC's replacement for the sprint: a shorter, more intense cycle measured in hours or days rather than weeks, sized to how quickly an AI can build.

Is AI-DLC the same as AIDD or AIDDLC?#

No. AIDD is a general term for building with AI tools. AIDDLC is a separate seven-phase standard from a different author, despite the near-identical name. AI-DLC is AWS's three-phase method.

Is AI-DLC open source?#

Its adaptive workflows are: AWS published them in November 2025 as awslabs/aidlc-workflows, under the MIT-0 licence.

Apply it in your role#

Across the enterprise. AI-DLC scales across teams when the context artefacts are standard and the mob rituals are timeboxed. Add what the method leaves to you: the bar per slice and the authority budget.

The ten-minute workflow. Prepare a Mob Elaboration session so the room's time goes on decisions:

Here is an intent: <one paragraph>. Act as the facilitator of a Mob Elaboration session. Propose the
units of work, then list the clarifying questions the team must answer before building, ordered by
how much of the plan each answer changes. For each question, say who in the room should answer it.

Sources and credits#