AI-Assisted Development
AI tools can accelerate implementation, but they need project context and boundaries.
Give AI high-quality context
Before asking for code, provide:
- the exact files to touch,
- architectural constraints (app boundaries under
apps/), - the expected user outcome,
- and non-goals.
Better context produces smaller, safer patches.
Preferred prompt shape
Use prompts like:
- Goal: what user job should work after the change.
- Scope: allowed files/folders.
- Constraints: coding style, architecture, no breaking changes.
- Acceptance criteria: what must be true when done.
Guardrails to enforce
- Keep business logic out of templates.
- Don’t mix API and page concerns in the same module.
- Use background tasks for long-running work.
- Preserve existing env variable names unless migration is explicit.
- Avoid speculative dependency additions.
Review checklist for AI-generated patches
- Does code follow the repository boundaries?
- Are failure paths handled and logged?
- Are new settings/config documented?
- Are migrations/tests included when required?
- Can another dev understand the change without extra context?
Practical workflow
- Ask AI for a small first patch.
- Review and run checks.
- Iterate in narrow increments.
- Update docs before merge.
Small iterations beat one giant AI-generated diff every time.