15.10.2025ยท00:06:53

Cursor's Plan Mode CHANGES Everything

Cursor just dropped Plan Mode, and it completely changes how you approach AI-assisted development. This new feature stops AI from making random changes to your codebase and instead creates a detailed, step-by-step plan before writing a single line of code. The analysis looks at how Plan Mode compares to regular Agent Mode, why it prevents "shoveling snow while it's still snowing," and how it automatically generates MD files you can reference later. Plus, the game-changing feature where Cursor actually asks clarifying questions before starting. โ€” TIMESTAMPS: 0:00 - Plan Mode Finally Lands 1:20 - The "6 Month Project" Reality Check 2:04 - Learning Complex Codebases Made Easy 3:23 - Automatic MD File Generation 5:15 - The Clarifying Questions Game-Changer โ“ 6:42 - What is "CMD + AI"? โ€” LINKS & RESOURCES: ๐Ÿ“ท Follow on Instagram: https://www.instagram.com/cmdaishow โ€” #Cursor #AICoding #PlanMode #DeveloperTools #Programming #AIDevelopment #CodeGeneration #Productivity #TechTools #SoftwareDevelopment

Transcript
Stop raw dog in your AI development. Plan mode is finally available in cursor and I am here for it. It is exactly what it says on the tim. I think they introduce a few little nicities here, but if you're not familiar with clawed code, which already has a plan mode, you shift tab, which is the same thing here. For me, this is the recommended way to build features. I think that the idea of a plan mode is when you when you send a agent off to do some coding, it's kind of figuring it out in real time. It kind of broadly looks at everything, but it's it's almost like shoveling snow while it's still snowing. It's making changes. Whereas this will just look at everything and plan it out. And if you can combine this with some MCP servers, if you can combine it with some uh level of like thinking mode and whatever, then you'll have a plan that it outputs to a MD file that one you can reference later on. I often do that. I'll make an MD file so that I can push it up and then I'll do that in different branch later on or whatever. Um but yeah, it will give you a plan before then you start executing it. And what of what often happens, it breaks into phases for some reason tells you it's going to be like this is week one and this is going to take a week to do and this is going to be three weeks. Like it still uses antiquated um what do you call Yeah. So, but I literally built a six-month project in a day. Like what do you mean you aren't you going to oneshot it for me like in 30 seconds? What do you mean six month? Yeah. Okay. So, If I say it in a different way because I noticed this but I haven't used it but so the the way I see it essentially today when you use cursor you have the option to chat or like use the agent. So the agent does something out of like just right away and it might break things. You don't want that. And the chat I noticed that I go back saying okay the things that you said should be done now do them. So I'm going like in this mode that should have a in between mode and this is exactly that that is kind of like chatting with the AI but what you get is not just a chat what you get is a plan that the AI can execute on in a step-by-step way. And this gets me excited because last week I started with um like a GitHub repo from somebody from the Webflow community to to essentially learn what this guy is doing like some crazy [ __ ] with animations and I want to you know get in and learn it. So I I'm trying to I I was trying to learn learn that using cursor and it was not doing what I wanted because I wanted to give me a plan to look at the entire doc the readme file everything and give me a plan how I can use this structure and start like building piece by piece but it the chat mode was not good because with the chat mode it gives you a plan but it's it's in a chat And you can't go you can't say okay number one is done let's go to the next one but you could you could say output this plan to a MD file but now we can now we can that's what I'm saying you couldn't do that before well I didn't know how to do it right you didn't think you just didn't think to ask just help put this plan think for me but but you know you again you're the master of your own destiny so just correct me if I'm wrong but it sounds like you want to implement this like you want to write it. You don't want the agent to write this code. Yes. In this case, your plan. Yeah. Yeah. I mean, in this case, I want Yeah. In this case, uh this repo is about me learning. It's not a framework, but close to like a yeah like a mini framework how to do like cursor and g like not cursor how to do like gap and uh 3JS and this kind of things inside of web flow. Yeah. with you know with this framework and I don't know a lot of these things so I have to start from the very basis but I want to do it following like this guy's experience I mean there's nothing stopping you going into chat GBT and again master your own destiny tell it that you want to implement it and that you it needs to tell you what code to add and why you're adding it and what it does right so you need to tell it that there's nothing stopping you doing this inside of chat GBT but the benefit here being inside a cursor is that has context to your codebase. So this is how to implement it in your files that you already have. So this is this is great. Um you know I always struggle to understand why this is any different to for me this is just a a system prompt a a ch you change the system prompt. I don't know why this would be any different to being in edit mode but you say do not make any changes just make a plan. I I don't know what the difference would be. It would be interesting to find that out. But it's nice just to know because I've, you know, I guess in some instances I have had it just start working on stuff. No, no, no, don't do anything like I want you to just whatever. Yeah. That all being said, you got to tell it that you want to implement it, but it will just Yeah, it will just look at all of your files and and make a comprehensive plan for you to do later on. And this one automatically makes an MD file for you. So I often have to ask claw code to make an MD file for me to output it to it. It just creates a plan and that's it. Ask me if it wants to start implementing it. The other one is um it'll I don't know whether you've noticed this, but you it'll actually autorecommend itself. So if you're writing something that sounds like you're asking for a plan, it will say, "Do you want to use plan mode?" So that's quite good. That's quite nice. Um and then it always comes back with questions as well. So that I like that feature. oftentimes claw code almost just never come back with questions trying to like clarify things. Maybe I should add that into my root memory file to say anytime I'm in plan mode, ask me clarifying questions because yeah, it's um that's a nice feature that I saw on this as well. So yeah, pretty good. Yeah, and I would recommend it to everyone building out big big features like just take a moment to plan it. Um, take a step back. Just be methodical. Be I know it's we're so eager to like [ __ ] just build build and like work fast. Like take your foot off the gas. Just be a controlled, organized freak when it comes to AI coding. This is part of a larger conversation on my show, Command AI, which we stream live every single week. We discuss the news and all things related to AI in the world of design and web. Catch us next week and join in the banter.