Most teams “use AI” by giving developers a coding assistant and keeping the rest of the process the same: two-week sprints, long planning meetings, hand-offs between roles. The gains are real but small. The AI-Driven Development Lifecycle (AI-DLC) takes a different view. If AI can do much of the planning, design and building, the process itself should change.
Why adding AI to agile isn't enough
Scrum and similar frameworks were designed around what human teams can do: how much work fits in a sprint, how often people can meet, and how long hand-offs take. When an agent can draft requirements in minutes and implement a story in an hour, those rhythms become the bottleneck. Teams end up waiting for the next ceremony, not for the work.
The opposite mistake is just as common: letting AI run without structure, with no shared context, no validation, and code nobody fully understands. AI-DLC tries to keep AI's speed and human judgement.
What AI-DLC is
AI-DLC was introduced by Raja SP and his team at AWS in July 2025, and AWS open-sourced a set of adaptive AI-DLC workflows in November 2025. Its core idea is simple:
AI drives, humans steer. AI drafts plans, asks clarifying questions and proposes solutions. People validate, decide and stay accountable.
Each phase produces artefacts that become the context for the next phase, so the AI's suggestions get better informed as the project moves along.
The three phases
Inception
Goal: turn business intent into clear requirements and Units of Work.
Ritual: Mob Elaboration. The product owner, developers, QA and other stakeholders meet with the AI. The AI asks questions (who are the users, what does success look like, what are the constraints?) and proposes user stories and units. The whole group validates them in real time.
Construction
Goal: turn validated units into tested, working software.
Ritual: Mob Construction. Using the Inception context, the AI proposes architecture, domain models, code and tests. The team settles technical decisions as they come up, and every output is reviewed before it's accepted.
Operations
Goal: deploy, observe and improve in production.
AI helps with infrastructure as code, deployments, monitoring and incident analysis. Production learning flows back into the next Inception.
New vocabulary
| Traditional agile | AI-DLC | What changes |
|---|---|---|
| Sprint (1–4 weeks) | Bolt | A short, intense cycle measured in hours or days |
| Epic | Unit of Work | A cohesive, loosely-coupled piece of value that can be built and deployed independently |
| Backlog refinement | Mob Elaboration | AI drafts, and the whole team validates live, in one session |
| Individual coding + PR | Mob Construction | AI implements; the team makes the technical calls together |
| Docs written afterwards (if at all) | Persistent context | Every decision and artefact is saved and reused by later phases |
Adaptive workflows: rigour that matches the risk
A sensible version of AI-DLC doesn't run every step for every change. AWS's open-source workflows let each phase decide how deep to go based on the work. A typo fix skips most of Inception. A new payments capability gets full requirements analysis, architecture review and testing. The guiding principles include context awareness, risk-based execution, question-driven elaboration and keeping the user in control.
A Bolt, hour by hour
Here's an illustrative two-day Bolt for adding online leave approval to an HR portal:
Mob Elaboration (90 min). HR head, a manager, two engineers and the AI agent. The AI asks a dozen or so clarifying questions; the team settles half-day leave, delegation and holiday rules. Output: approved spec and three Units of Work.
Plan & design. The AI proposes the data model, API contracts and UI flow. The architect changes one decision (notifications through the existing RabbitMQ exchange instead of a new service).
Mob Construction. Agents implement unit 1 test-first in .NET 10 and React. Engineers review each pull request and answer the agent's questions.
Units 2–3 + Mob Testing. QA and the product owner try real scenarios against the acceptance criteria; any failing edge cases go back to the AI and are fixed the same day.
Operations. The Azure DevOps pipeline deploys to staging, then production behind a feature flag. Monitoring and alerts are added from the spec's non-functional requirements.
Illustrative example. Real Bolts vary with scope, team and risk.
Where spec-driven tools fit
AI-DLC is a methodology. You still need tools that capture and organise the artefacts. In our projects:
- BMAD Method structures Inception for larger initiatives (brief, PRD, architecture, stories).
- GitHub Spec-Kit turns a Unit of Work into spec → plan → tasks for Construction. See our Spec-Kit guide.
- OpenSpec handles changes to existing systems as reviewable proposals. See our OpenSpec & BMAD guide.
- Superpowers skills make coding agents brainstorm, plan, work test-first and request review.
- Claude Code, Cursor and GitHub Copilot do the hands-on implementation.
- Azure DevOps / GitHub Actions run the quality gates and deployments in Operations.
What humans still own
- Intent and priorities: what's worth building and why.
- Validation: approving requirements, designs and code before they move on.
- Trade-offs: cost, security, compliance and architecture decisions.
- Accountability: an engineer signs off every change that reaches production.
Getting started checklist
- Pick one pilot Unit of WorkValuable but low risk, ideally a feature on an existing product.
- Choose your artefact toolingSpec-Kit for new features, OpenSpec for brownfield changes, BMAD if you need full planning.
- Book a 90-minute Mob ElaborationGet the decision-makers in the room with the AI, not just developers.
- Run a 1–3 day BoltMob Construction with mandatory review, then deploy behind a feature flag.
- Measure and adjustTrack lead time, rework and defects against your last similar feature.
AI-DLC at MD Technosoft
We run AI-DLC on client projects across .NET, React/Angular, Flutter and cloud, and we help in-house teams adopt it. Our Forward Deployed Engineers can lead your first Bolts inside your team, so the practice stays with you afterwards. Learn more about our AI-DLC service.
