@Marie Monet to be honest, if I will go with only ONE choice, I will go with claude code. Try also to take the Max 5x and see how it will change your life, and if you can copy lovable repos to local and try to finish them using Claude Code. Then, you can even cancel lovable, you will also notice you won't need it anymore at one point, but do the migration slowly until you feel sure you got this right with Claude Code. Same thing I did during my migration from cursor. Claude Code also will cut your N8N workflow creation time to 90%, because you can connect an mcp which has direct connection to your whole N8N server. Back to your main question. The best way I tackle forgetfulness is by creating a fundament and a base the ai can always go back to. Meaning, before you even start you need to emphasize so much on the plan (even Sabrina was mentioning this), also create a mock if you can which will have everything for you before you even start. And, every time the AI deviates redirect it to the plan, or mock, this will give it complete autonomy and understanding of what is going on.. And, to create a more robust guard to also have more info. you can also create a progress.md file which will always be updated based on what have been changed, and just ensure it is always up to date. Finally, you can also use mcps like gsd or subtasks in Claude Code which will have these things internally added.