Overview
This video gives a complete walkthrough of Equinox’s AI-First software development life cycle, showing how AI supports every phase from planning and design through implementation, testing and delivery. The structure of the SDLC stays the same, while AI-driven prompts, tools and automation speed it up. Most initial code is now AI-assisted, so developers focus on reviewing logic, refining implementations and keeping to the architectural standards.
AI-First SDLC – Planning
Planning starts with user stories. Predefined guardrail prompts, customized for the project, are applied to give the AI’s output structure, boundaries and clarity, and the prompt templates have been refined over time so the results are consistent. The process distinguishes AI-driven activities from manual ones. In planning, prompts derive the initial design inputs that guide the design and implementation steps.
AI-First SDLC – Design
Design uses a structured two-step prompting approach. First, the context is set with the architectural landscape, domain constraints and technical guidelines. Then the user stories are added, covering functional and non-functional expectations. AI then produces design decisions, implementation approaches and key architectural considerations such as data models, API flows, event handling and security impacts, and refines them iteratively. The sample design document is then checked manually for accuracy, feasibility and alignment with the architectural standards. The design goes through three levels of review: an AI-assisted review with predefined prompts, iterative regeneration with Copilot prompts, and a thorough manual review by architects, tech leads and sometimes QAs. The final design document supports both development and test case generation.
AI-First SDLC – Implementation and Unit Testing
Implementation uses the Node.js stack, with architectural guardrails, domain-driven design principles and structured system prompts guiding the coding. The AI is given three inputs, the system prompt, the user story and the design document, and it produces a development plan, implementation steps and initial code. The plan is reviewed with AI review prompts and regenerated as needed before the code is produced. Developers then work interactively with Copilot to debug and resolve compilation issues until the code is stable. For unit testing, the implementation code and the user stories generate a test plan and test cases, which are reviewed, refined and executed.
AI-First SDLC – Functional Testing and Automation
QA teams receive the user stories, design documents, implementation details and unit test scenarios, and use AI prompts to generate functional test cases. In the 9.0 release, the approach shifts left: developers generate the unit tests, QA reviews them, and a system prompt takes the user story and the unit tests to generate the functional test cases. For APIs, the schema details are provided too. The test cases go through a two-step review, a manual QA review and then an AI review prompt, and are saved in the code base as markdown files. For automation in 9X, the team built an API testing framework on Playwright and is moving away from Ready API. A second system prompt takes a JSON template and the functional test cases and generates the JSON files the framework needs. Each prompt states which test cases to generate, which to skip and which rules to follow.
AI-First SDLC – Common Questions
The video answers a few common questions. Development uses GitHub Copilot inside Visual Studio Code. There is no strict prompting method or coding style, because the guidelines are a foundation and not a restriction. If a user story changes midway, the context is reset, the updated story is provided and the AI updates only the impacted part. Markdown test cases are converted to Excel with AI and imported into Zephyr. The approach works for both mono-repo and multi-repo setups. Because AI handles generation end to end, traditional story sizing such as small, medium and large becomes less relevant. A sample design for updating a user profile, which triggers an event for activity logging, shows the API design, data models, synchronous or asynchronous handling and security considerations such as JWT for authentication.
Timeline
- 00:04 – Overview: AI across every phase of the SDLC
- 00:48 – AI-assisted code generation and the developer’s focus
- 01:03 – Planning: user stories and guardrail prompts
- 01:47 – Design: two-step prompting with context and user stories
- 02:35 – Sample design document and manual review
- 03:14 – Three levels of design review
- 03:40 – Implementation: system prompt, user story and design document as inputs
- 04:18 – Development plan review, code generation and working with Copilot
- 04:38 – Unit testing: test plan and test cases
- 05:01 – Functional testing: QA inputs and AI-generated test cases
- 05:27 – Common questions: Copilot, prompting style and changing user stories
- 06:29 – Sample design: user profile update with an activity logging event
- 07:09 – Functional testing in 9.0: shift left and story to test case prompting
- 08:08 – Two-step review and markdown test cases
- 08:51 – Test automation: Playwright API framework and test case to JSON prompting
- 09:55 – More questions: Zephyr upload, AI models, repo setups and story sizing