From Vague Idea to Testable User Stories in Minutes: AI-DLC Inception in Amazon Quick Desktop
How a custom AI skill turns the hardest part of building software - defining what to build - into a structured, repeatable conversation
The Problem Every PM Knows Too Well
You have a new feature idea. Maybe it came from a customer conversation, a leadership review, or a late-night insight. The concept is clear in your head, but between that spark and a set of well-defined, testable user stories lies hours (sometimes days) of work: stakeholder alignment, edge-case discovery, acceptance criteria writing, architecture decomposition, and team planning.
What if you could compress that entire journey into a structured conversation with an AI that acts as your expert PM - one that proposes, but never decides for you?
That's exactly what AI-DLC Inception does inside Amazon Quick Desktop .
What Is AI-DLC?
AI-DLC (AI-Driven Development Lifecycle) is a methodology that brings engineering discipline to AI-assisted software development. It has three phases:
- Inception - Defining intent, requirements, and architectural direction
- Construction - Building features via short "Bolts" iterations
- Operations - Production monitoring and feedback loops
AI-DLC is tool-agnostic. It works with Kiro, Cursor, Claude Code, GitHub Copilot, or any AI coding assistant. But when paired with Amazon Quick Desktop, the Inception phase becomes something you can trigger with a single command.
Why Amazon Quick Desktop?
A methodology is only as good as the tool that runs it. For a PM running AI-DLC Inception, the tool needs to do more than just chat. Here's what Amazon Quick Desktop brings to the table:
- Connect to your existing tools - Slack, Outlook, Gmail, Salesforce, local files. As a PM, your requirements context is scattered across all of these. When Quick runs Inception, it can pull from any of them - a Slack thread where the customer described the problem, an email with stakeholder priorities, a spreadsheet with constraints.
- Remember your projects and people - No starting from scratch every session. Quick builds an understanding of your work that gets smarter every day. So when you say "write stories for the dashboard project," it already knows the stakeholders, the prior decisions, and the technical context.
- Work with any file on your machine - Inception produces Markdown files and folder structures directly on your filesystem. You can point Quick at existing docs, PRDs, or competitor analyses and it incorporates them into the stories it writes.
- Produce real deliverables, not chat responses - The output isn't ephemeral. It's
user_stories.md,units_plan.md, and individual unit files saved to a timestamped project folder on your machine. Version-controllable. Reviewable in PRs. Ready to hand off. - Custom skills turn any workflow into a command - This is the key. Quick Desktop lets you create reusable skills - structured instructions + tool bindings + trigger phrases - that turn any repeatable PM workflow into a single command. AI-DLC Inception is exactly this: a set of skills you invoke in conversation. And you can build more for your own workflows (sprint planning, stakeholder updates, competitive analysis - anything that follows a repeatable pattern).
Amazon Quick Desktop is available for MacOS and Windows.
All of these capabilities work together during Inception. Quick uses its tool connections to gather context, its memory to understand your project, its file access to write the deliverables, and its skill system to run the structured workflow - all in one conversation, all from your desktop.
How Inception Works Inside Quick Desktop
I built a custom skill for PMs that automates the Inception phase. Here's the workflow:
Step 1: You Provide a Use Case
It can be as simple as: "Build a customer feedback portal where users can submit feature requests and upvote existing ones."
That's it. One sentence. Quick takes it from there.
Step 2: Quick Acts as Your Expert PM
The AI shifts into PM mode. It doesn't just start writing stories - it plans the work first. You get a structured plan with checkboxes showing exactly what it intends to do:
- Identify the key actors and personas
- Map out core user journeys
- Flag ambiguities that need your input
- Propose story groupings
Step 3: You Approve Before It Proceeds
This is the critical principle: the AI proposes, the human approves.
Quick surfaces anything that needs your judgment - scope boundaries, priority trade-offs, technical assumptions - and waits for your confirmation. It never makes critical decisions autonomously.
Step 4: Complete User Stories with Rich Acceptance Criteria
Once approved, Quick delivers fully formed user stories, each with three layers of acceptance criteria:
- Functional requirements - What the system must do
- UI/UX requirements - How it should look and feel
- Non-functional requirements - Performance, security, accessibility constraints
Every story follows a consistent format. Every acceptance criterion is testable. No more ambiguous "the system should be fast" requirements.
Step 5: Architect Mode - Units for Parallel Development
After stories are complete, Quick switches to architect mode. It analyzes dependencies between stories and groups them into loosely coupled units - logical system areas that can be developed in parallel by different team members or AI agents.
Each unit is independent enough to be built, tested, and deployed without blocking other units. This is what enables true parallel team development, whether your "team" is humans, AI agents, or both.
The Key Design Principles
1. Human-in-the-Loop, Always
The skill is deliberately designed so that the AI never makes a scope decision without explicit approval. If it's uncertain about a requirement, it asks. If there are multiple valid approaches, it presents options.
2. Structured Output, Not Free-Form Text
Every output follows a defined format. Stories are consistent. Criteria are testable. Units have clear boundaries. This isn't a chatbot generating prose - it's an engineered workflow producing engineering artifacts.
3. Repeatable Process
Run it on ten different use cases and you get ten consistently structured outputs. The quality doesn't depend on whether you remembered to ask the right questions. The skill's internal plan ensures comprehensive coverage every time.
4. Tool-Agnostic Downstream
The user stories and units AI-DLC Inception produces can feed into any development workflow. Hand them to a human team. Feed them into Kiro's spec-driven development. Use them as prompts for AI coding agents. The output is the universal starting point.
Why This Matters for Builders
The hardest part of building software isn't writing code - it's deciding what to build. Poor requirements lead to rework. Vague stories lead to scope creep. Missing edge cases lead to production incidents.
AI-DLC Inception doesn't replace PM judgment. It amplifies it. It handles the mechanical work - formatting, consistency, coverage checking, dependency analysis - while keeping every strategic decision in human hands.
For PMs , this means:
- Less time on documentation mechanics, more time on product thinking
- Consistent quality regardless of time pressure or cognitive load
- Faster feedback loops - you can iterate on stories with stakeholders the same day you conceive the feature
- Ready-to-execute artifacts that teams (human or AI) can pick up immediately
The Skills I Built
I created three composable skills in Amazon Quick Desktop that implement the AI-DLC Inception phase. Creating them is easy - in the Capabilities panel, you write instructions in a structured SKILL.md format, attach the tools the AI needs, set a trigger phrase, and it's live. No code required.
The three skills:
- AI-DLC Inception Stories (4 tools: Folder Create, File Write, File Edit, File Read) - Acts as expert PM. Plans the work, asks for approval, writes complete user stories with functional, UI/UX, and non-functional acceptance criteria.
- AI-DLC Inception Units (5 tools) - Acts as expert architect. Reads the completed stories, analyzes dependencies, and groups them into loosely coupled units for parallel development.
- AI-DLC Standards - Shared format steering document loaded by both skills for consistent output across runs.
They're composable - Stories runs first, Units reads from Stories' output. Each can also run standalone.
See It in Action
Here's a real run where I said: "Create user stories for a note-taking app."
Quick proposes a plan and asks for approval:
After approval, it delivers the complete user stories:
Then in architect mode, it decomposes stories into parallel units:
Final output - clean structure ready for development:
What Happens Next: From Inception to Construction and Beyond
Inception gives you the what. The next question is: how do you build it?
The artifacts AI-DLC Inception produces - user stories with testable acceptance criteria, units with clear boundaries and dependencies - are designed to flow directly into the next phases of the AI-DLC lifecycle:
Construction Phase
Each unit file is a self-contained brief that any developer or AI coding agent can pick up. You can:
- Feed a unit directly to Kiro - Its spec-driven development workflow reads your stories and acceptance criteria as the spec, then builds and tests against them.
- Use it with Cursor or Claude Code - Paste or reference the unit file as context. The structured format (stories + acceptance criteria + backend/frontend breakdown) gives the AI exactly what it needs to generate code without ambiguity.
- Hand it to a human team - The Markdown files work as traditional engineering specs. No translation needed. Developers read the acceptance criteria and know exactly what "done" looks like.
- Run multiple units in parallel - Because units are loosely coupled, different team members (or AI agents) can build them simultaneously without blocking each other.
Operations Phase
The acceptance criteria from Inception become your test contracts in production:
- Non-functional criteria (latency targets, scale limits) become monitoring thresholds and alerts
- Functional criteria become integration test cases
- The unit boundaries inform your service boundaries and deployment topology
The key insight: AI-DLC is a continuous flow, not isolated phases. Inception artifacts feed Construction, Construction artifacts feed Operations, and Operations feedback loops inform the next Inception cycle. Whether you use Quick Desktop skills, Kiro, Cursor, GitHub Copilot, or plain terminal workflows - the structured artifacts from Inception are the universal handoff format that works with any tool in your stack.
Conclusion
We're past the point of asking whether AI should be involved in software development. The real question is: how do we keep humans in control while letting AI handle the mechanical parts?
AI-DLC Inception answers that question for the requirements phase. The AI proposes, the human approves. Every critical decision stays with you. And what used to take days now takes minutes.
The hardest part of building software just became a structured, repeatable conversation.
Nitin Eusebius is a builder at AWS who focuses on making AI-assisted development more disciplined and repeatable.
#AI-DLC #ProductManagement #HumanInTheLoop #RequirementsEngineering #KiroEnjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article