Defra Logo Defra Logo

Defra AI in the SDLC Playbook

This playbook is a living document that aims to provide guidance on best practices for ways of working, tooling, and context-specific use cases for the use of AI in the Software Development Lifecycle (SDLC) tailored to Defra's unique needs and challenges.

Development

This section outlines how to develop new features. For test-first development, see Test Driven Development.

Workflow

Prerequisites

  1. Clear requirements: Write detailed feature requirements with defined scope, as outlined in Product Requirements
  2. AI rules: Set up project Rules for AI to maintain consistent standards, patterns and conventions across your codebase
  3. Capable model: Use the most capable large language model (LLM) available for your task to get quality results

1. Create a new git branch

Create a new git branch for your feature.

2. Prompt the coding assistant

  • Use an AI Coding Assistant (AICA) in Agent mode
  • Use prompt-new-feature-story and reference your requirements files directly with the Coding Assistant Agent
  • Generate the initial code
  • Accept changes, then review them in the git diff viewer

3. Test and iterate manually

  • Test the generated feature manually to check it works correctly
  • Give error messages back to the model for quick fixes
  • Review the code to make sure it meets your expectations once it’s working

4. Create automated tests

  • Start a new agent conversation
  • Use testing prompts from your prompt library to generate tests
  • Accept changes, then review in the git diff viewer
  • Review the tests carefully to make sure they meet your expectations once they’re working

5. Refactor

Follow-up tasks

  • Create a merge request for peer review
  • Update documentation using prompt-add-update-documentation
  • Update AI rules and prompt library based on what you’ve learned

Guidelines

Get consistent results: Follow the four pillars by using AI rules, prompts and clear product requirements together.

Review everything generated: Review and refine everything the agent generates.

Debug with logs: Include detailed logs when developing. Give error messages back to the model for faster fixes.

Avoid the “Doom Loop”: If you need multiple iterations, your prompt was unclear. Revert changes, refine your prompt, and restart.

Match scope to complexity: Give the AI more specific tasks as codebases grow.

Next -> Test Driven Development

Support links

All content is available under the Open Government Licence v3.0, except where otherwise stated