Wrote a while ago about using “plan mode” in v0. Check out the article originally posted in LinkedIn or you can read it here below.
From Prompt to Plan: Building with Intention in v0
In the Vercel Community channels, we tend to see a reoccurring theme discussing why Plan Mode in v0 by Vercel with any of the v0 agents is a must and a definite game changer. So I wanted to give my two cents and experience on this.
Getting the best results from v0 by Vercel isn’t about letting the AI agents run wild. It comes down to many things, two of those being: how clearly you instruct it (aka prompting), and how well you plan before anything is built. Prompting properly matters always (read more on that here), whether that’s custom instructions or using “@” mentions to apply specific behaviors, but planning matters just as much. That’s where I think Plan Mode truly earns its place.
Plan Mode fundamentally changes the workflow in the best ways possible. Instead of jumping straight into design + code, it forces a pause. The AI must think, explain, and align with your intent before touching a single file. That shift alone eliminates a huge class of accidental changes and wasted effort.
Plan Mode selected in v0
You can enable Plan Mode either by selecting the preset instruction (image above) or by explicitly calling it with “@Plan Mode” in your prompt. That mention is the contract which tells the agent to slow down and plan first. It also helps to speak to the agent in a way that actively reinforces this behavior. As part of step one, you can end your prompt with something like “please confirm your plan before proceeding.” That small nudge consistently keeps the agent in planning mode instead of drifting into execution. Its a small step which I use a lot and find it saves more time in the long run. Whats even better is you can now have Plan Mode ask you clarifying questions at first prompt! See more on that here.
For example, say you want to add a persistent dark mode toggle to an existing React app. You might start with a prompt like, “@Plan Mode add a dark mode toggle to the app that persists the user’s preference across sessions. Please confirm your plan before proceeding.”
Instead of generating design + code immediately, the agent responds with its understanding and, you guessed it, a plan. It outlines which components will change, how states will be stored, whether localStorage or system preferences should be used, how Tailwind’s dark mode will be configured, and what edge cases (if any) need to be handled. At this point, nothing has been written. You’re simply reviewing intent, not output.
This is where you can then refine the plan if you feel its not fully there or something is missing. Or maybe, its just sounds like a terrible plan to you and you need to pivot somehow, or create a new plan altogether. You could then say to the agent, “prioritize {prefers-color-scheme}, make the toggle optional, and avoid introducing new dependencies.” The agent updates the plan accordingly and only once you approve, does it move on to implementation.
That approval gate is the real power of Plan Mode in my opinion. You see exactly how the AI intends to solve the problem before it spends credits or touches your codebase. Errors drop, rework drops, and credits don’t feel “wasted”. The resulting code fits your architecture instead of fighting it, and all while at a low credit cost!
In practice, Plan Mode feels less like prompting an AI and more like reviewing a thoughtful technical proposal before green lighting implementation. It’s like a design review, but for code. Strategy first, execution second. The kind of discipline that quietly saves time, money, and future frustration.
So, if you haven’t tried Plan Mode, or maybe haven’t explored it as much, hop into v0 by Vercel today and get to planning! (pun very much intended 
)