What are you trying to build, change, or untangle?
Practical guides for the questions behind the architecture terms: where to start, why fixes keep breaking things, whether to rewrite, and how to give your coding agent a direction it can build from.
Start with the problem you recognize
Choose the page that matches what you are trying to build, change, or understand.
I am not sure what to call the problem
Find your situation without needing architecture vocabulary.
Start hereYou need a stable starting point
Start with the decisions a coding agent must not invent as the project grows.
Start hereThe app feels messy
Start with the warning signs that a working product is drifting out of shape.
Start hereThe codebase feels risky
Start with a bounded recovery route instead of another blind feature request.
Start hereThe app disagrees with itself
Start by finding the one record or workflow that should govern a disputed decision.
Start herePlanning keeps going nowhere
Start by separating useful exploration from decisions that need explicit ownership.
Start herePublished pages
My AI-built app works, but I am afraid to change it. What now?
Do not start by rewriting a working app. First identify one important user journey, what currently owns its state, and what must remain true when it changes. Then make and verify one bounded change.
Read guideGuideAI keeps fixing one thing and breaking another: how to stop the loop
When each fix breaks something else, stop stacking patches. Reproduce one failure, identify the shared state or boundary involved, and require the repair to preserve the working journeys around it.
Read guideGuideI have an app idea. What should I give an AI coding agent before it builds?
Give the agent more than a feature list: define the first useful user outcome, the product rules that govern it, the important unknowns, and a complete first build boundary. Technical choices should support that direction, not silently invent it.
Read guideGuideHow to add a feature to an existing AI-built app without redesigning everything
Treat a feature request as a change to an existing system. State the new user outcome, inspect the boundaries it touches, name what must survive, and verify the old journey as well as the new one.
Read guideGuideShould I rewrite my AI-built app, refactor it, or keep building?
Do not choose a rewrite from the size of the mess alone. Identify what works, what owns important state, what users rely on, and whether a bounded repair can restore a complete journey. Rebuild only the boundary the evidence justifies.
Read guideGuideMy coding agent keeps forgetting decisions and rebuilding the wrong thing
A longer chat history is not a reliable product authority. Keep the governing product decisions, the current implementation state, and the next bounded outcome distinct, and make deliberate changes propagate between them.
Read guideGuideMy AI-generated app looks finished. How do I tell if it is ready to launch?
A polished screen and a passing build are not launch evidence. Verify the complete promised journey in the deployed environment, including saved state, access control, interruptions, and the ability to recover when something fails.
Read guideGuideApp architecture vs technical blueprint vs implementation plan: what do I need?
Architecture defines the system and its rules. The technical blueprint explains how those rules are implemented. The implementation plan orders complete outcomes and their verification. You need enough of each to build your actual product, not three large documents for their own sake.
Read guideGuideHow AI-built apps drift out of shape
AI-built apps rarely fail because one prompt was bad. They fail because product truth, code truth, and the build workflow split apart.
Read guideGuideWhen your AI-built app has three different sources of truth
Many ?random bugs? are really authority conflicts: the interface, the database, and the AI instructions each believe they own the same decision.
Read guideGuideHow to continue a messy Cursor or Claude Code project without starting over
The safest next move is rarely another feature. First separate what still carries product truth from what only carries implementation noise.
Read guideGuideWhy AI planning chats fail on real apps
A planning chat can sound smart while never forcing the decisions that keep a real app coherent.
Read guideGuideHow to recover a Cursor-built app without starting over
Recovery gets easier once you stop arguing in absolutes and separate product truth from implementation noise.
Read guideGuideArchitecture before vibe coding: how to build an AI app without drift
You do not need to know every framework before building with AI. You do need to settle the decisions a coding agent must not invent for you.
Read guide