Find the detailed version of this roadmap along with other similar roadmaps
AI Tools can Help with Planning
For example if you are using "Claude Code", tell the tool what you are trying to build, ask it to help refine the idea, establish the different phases, and once done ask it to document everything in a document so that you can refer to that when actually building the product.
Plan what you need to develop (MVP, Different Phases)
Work step by step rather than trying to build everything at once
Illustrate AI with examples (mockups, code samples, images)
Implement Spec-Driven Development (SDD)
Establishing Standards Early is Important
When you're starting out, carefully review the AI's initial outputs - coding patterns, styles, architecture decisions. If you don't catch bad habits early, the AI will reinforce them with every iteration, and they'll compound fast.
Pick a popular tech stack rather than new/niche ones
If you have style/coding preferences, document them for AI
Ask AI to keep the code modular and aim for smaller modules/files
Regularly ask the AI to review and refactor the codebase
Never Hardcode your Secrets
Use skills created by others
Never let the AI put passwords, API keys, or tokens directly in the code. If you see it doing this, stop it and ask it to use environment variables instead.
Explicitly ask AI to perform a security audit of the application
Never hardcode or credentials; use env variables instead
Force refactoring sessions regularly
AI will take the path of least resistance (appending code, growing files, skipping cleanup). Periodically ask it to step back and refactor: break things into smaller modules, remove dead code, improve performance.
Ask for one task at a time rather than five different items.
Be specific about what you want, rather than high-level, vague instructions
Based on your previous coding sessions, tell AI what NOT to do
Keep Context Document Up to Date
Give AI mockups, reference files and material that can help it
If you see yourself repeating some instructions, document the instructions in the context document. One more thing that helps is: after a discussion session if you feel like AI can benefit from learnings in the session, ask it to update the context document (e.g. claude.md) with the learnings from the session.
Use "act as" framing when helpful (e.g. act as a UX researcher)
Regularly update your context document (e.g. CLAUDE.md)
Explicitly tell AI to "think" or "brainstorm" before complex problems
Leverage long context window when available and necessary
Clear context for better results and to save token costs. Whenver you are about to start a new/unrelated task, it's better to clear the context and start fresh.
If AI fails after 3 prompts, stop, and start a fresh chat
For unrelated tasks, proactively clean and start new sessions
Ask AI to use subagents, if possible
Let AI Debug, But Understand what Went Wrong
Prompt error messages and let AI do the rest
When something breaks, paste the error and relevant code and ask AI to explain the error. Don't just accept the fix — make sure you understand it. If you don't, ask AI it to explain in simple terms.
If errors persist, ask Al to create a list of possible causes
Tell AI to add logs to find the bug faster
Install and ask AI to use MCP (e.g. Playwright for browser), when possible
Commit Often, With Clear Messages
After each working feature or fix, make a commit. Ask AI to suggest a clear commit message that describes what changed and why. This gives you a safe checkpoint to roll back to if something goes wrong later.
Use `git commit` regularly (e.g. after every successful AI task)
Start each new feature with a clean Git slate
If you need to revert, use Git rather than AI native revert functionality
Ask AI to handle your Git and GitHub CLI tasks
Force AI To Test by Default
Ask AI to write tests (E2E tests can help build a stable product)
Most AI tools' default behavior produces implementation-first code with minimal test coverage. Whenever AI builds a feature, force it to write basic tests right away to ensure that bugs get caught early.
Consider Test-driven development (TDD)
When you find a bug, ask AI to write a breaking test and then fix
Once tests are in place, refactor regularly
Continue Learning with following relevant tracks