NEW Gemini CLI Update is INSANE!

AI Training 👉 https://sanny-recommends.com/learn-ai

Most people using AI for coding run into the same problem. You ask the AI to build something, it jumps straight into writing code, and at first it looks impressive. Then you run it and everything breaks. Files are wrong, logic is off, and you’re stuck fixing things that shouldn’t have gone wrong in the first place.

Google just introduced a solution to that problem with a major update to Gemini CLI called Plan Mode, and it completely changes how AI works with your code. Instead of rushing into execution, Gemini CLI now starts by thinking first, mapping everything out before touching a single file.

Gemini CLI itself is Google’s open-source AI tool that runs directly in your terminal. You can install it with one command and use it to read files, write code, run commands, and interact with your entire project. It runs on Gemini 3 and even has a free tier with up to 1000 requests per day, so you can use it without paying anything.

The big change is how Plan Mode works. When you activate it, Gemini CLI enters a strict read-only state. It physically cannot modify your files. Instead, it explores your codebase, analyzes your project, and builds a structured implementation plan before doing any work. This alone solves one of the biggest issues with AI coding tools, which is acting too fast without understanding the full context.

While in planning mode, the AI reads your files, searches your codebase, and even looks up relevant information if needed. It then creates a detailed step-by-step plan in markdown format. Before anything happens, you review that plan, edit it, and approve it. Only then does the AI move into execution.

Another powerful part of this update is how Gemini CLI asks questions during planning. If it reaches a decision point, it doesn’t guess. It stops and asks you directly. That means you’re guiding the architecture before any code is written, instead of fixing mistakes after the fact.

It also uses automatic model routing, which is a big upgrade most people won’t notice at first. During planning, it switches to a higher reasoning model for better decisions. Once you approve the plan, it switches to a faster model for execution. You get better thinking upfront and faster results during implementation without needing to configure anything.

You also get full control over the plan itself. You can open it in your own editor, rewrite steps, reorder tasks, and leave comments. Gemini CLI reads your changes and updates the plan accordingly. This turns the process into a true collaboration instead of just prompting and hoping.

Recent updates have made this even stronger. New subagents can explore your codebase more deeply during planning, annotation support allows you to leave feedback directly inside the plan, and new commands let you quickly copy or reuse plans in your workflow. Google is clearly focusing heavily on improving this feature.

The reason this matters is trust. Most developers hesitate to use AI on real projects because of the risk of breaking things. Plan Mode removes that risk by guaranteeing nothing changes until you approve it. You can see exactly what will happen before it happens.

This makes AI much more useful for complex tasks like refactoring large codebases, working with legacy systems, or building features that touch multiple files. Instead of reacting to mistakes, you’re preventing them entirely.

If you want to actually learn how to use tools like Gemini CLI, Claude Code, and other AI systems to automate your workflows and scale your business, the AI Profit Boardroom is where you should be. It’s built for people who want to go beyond just watching AI updates and actually implement real systems that save time and generate results.

AI Training 👉 https://sanny-recommends.com/learn-ai

6 comments