{"id":3639,"date":"2026-03-16T13:14:40","date_gmt":"2026-03-16T13:14:40","guid":{"rendered":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/2026\/03\/16\/gemini-cli-plan-mode-separates-thinking-from-doing-and-makes-read-only-the-default\/"},"modified":"2026-03-16T13:14:40","modified_gmt":"2026-03-16T13:14:40","slug":"gemini-cli-plan-mode-separates-thinking-from-doing-and-makes-read-only-the-default","status":"publish","type":"post","link":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/2026\/03\/16\/gemini-cli-plan-mode-separates-thinking-from-doing-and-makes-read-only-the-default\/","title":{"rendered":"Gemini CLI Plan Mode Separates Thinking From Doing \u2014 and Makes Read-Only the Default"},"content":{"rendered":"
\"\"<\/div>\n

\"\"<\/p>\n

The pattern across AI coding tools this week has been clear: the industry is building governance, review, and safety mechanisms as fast as it\u2019s building capabilities. Google\u2019s latest contribution is plan mode for Gemini CLI, announced March 11<\/a>, and now enabled by default for all users.<\/span><\/p>\n

Plan mode puts Gemini CLI in a read-only state where the agent can navigate your codebase, search for patterns, read documentation, and map dependencies \u2014 but it cannot modify any files except its own internal plans. The agent researches your request, asks clarifying questions, and proposes a strategy for your review before any code changes are made.<\/span><\/p>\n

The idea is simple: Think before you act. The implementation has some features that make it more interesting than it sounds.<\/span><\/p>\n

How it Works<\/b><\/h3>\n

Enter plan mode by typing <\/span>\/plan<\/span>, pressing Shift+Tab, or asking the agent to \u201cstart a plan for\u201d whatever you need. Gemini CLI restricts itself to read-only tools \u2014 <\/span>read_file<\/span>, <\/span>grep_search<\/span>, <\/span>glob<\/span> \u2014 and can use specialized sub-agents, such as the codebase investigator, to map system dependencies.<\/span><\/p>\n

The agent creates an implementation plan as a Markdown file. You can review it, edit it directly, or provide feedback in the conversation. When you approve, Gemini CLI switches to an edit-capable mode for implementation.<\/span><\/p>\n

Model routing adds an important dimension. In plan mode, Gemini CLI automatically routes to higher-reasoning Pro models \u2014 specifically Gemini 3.1 Pro \u2014 for architectural decisions. When it shifts to implementation, it routes to faster models. Strategy gets the reasoning model. Tactics get the speed model.<\/span><\/p>\n

The ask_user Tool<\/b><\/h3>\n

Plan mode introduces a new <\/span>ask_user<\/span> tool that changes the dynamic between the developer and agent. Instead of making assumptions about your intent, the agent can pause its research and ask targeted questions \u2014 present options, request clarification on an architectural choice, or ask where a hidden configuration file lives.<\/span><\/p>\n

This bidirectional communication during the planning phase means the plan that emerges actually reflects what you want, not what the model guessed you wanted. It\u2019s a direct response to one of the most common failure modes in AI-assisted development: an agent confidently implementing the wrong thing because it was never asked.<\/span><\/p>\n

Read-Only MCP Integration<\/b><\/h3>\n

Plan mode isn\u2019t limited to local files. It supports read-only MCP tools, which means the Gemini CLI can pull context from your entire developer stack during the planning phase \u2014 read a GitHub issue, inspect a Postgres schema, search Google Docs \u2014 all without risking any modification to your codebase or external systems.<\/span><\/p>\n

For DevOps teams, this is significant. Planning a database migration? The agent can read the current schema, check the issue tracker for related tickets, and review existing documentation before proposing an approach. All in read-only mode. The codebase stays untouched until you explicitly approve the plan and switch modes.<\/span><\/p>\n

Conductor: The Orchestration Layer<\/b><\/h3>\n

Plan mode becomes especially powerful with Conductor, the Gemini CLI extension for context-driven development. Conductor organizes work into \u201ctracks\u201d with written specifications and task-oriented plans stored as persistent Markdown files in your repository \u2014 not ephemeral chat logs.<\/span><\/p>\n

Conductor now leverages plan mode for research phases, performing exhaustive pre-flight checks with zero risk. It uses <\/span>ask_user<\/span> to confirm critical decisions at each milestone. The workflow follows a clear progression: context, spec and plan, then implement.<\/span><\/p>\n

Google is working on bringing Conductor into Gemini CLI as a built-in mode \u2014 a signal of how central the plan-first approach is becoming to their agent strategy.<\/span><\/p>\n

\u201cGoogle\u2019s Gemini CLI Plan Mode signals a shift in how AI coding agents are governed, moving approval control from autonomous execution to deliberate, human-confirmed workflows before any changes are applied. This positions Google to compete directly for enterprise adoption where deployment risk tolerance is low and audit requirements are non-negotiable,\u201d according to Mitch Ashley, <\/span>VP and practice lead for software lifecycle engineering at<\/span> The Futurum Group<\/span><\/a><\/p>\n

\u201cIn practice, teams evaluating agentic coding tools will treat plan-first execution as a baseline governance requirement. Vendors that treat autonomous execution as the default will face procurement friction as enterprise buyers require explicit control checkpoints before granting agents broader operational autonomy.\u201d<\/span><\/p>\n

Why This Matters for DevOps<\/b><\/h3>\n

Plan mode addresses a specific anxiety every team using AI coding agents has experienced: the agent that starts making changes before you\u2019ve agreed on an approach. Read-only exploration as the default flips the assumption from \u201cact first, review after\u201d to \u201cresearch first, act when approved.\u201d<\/span><\/p>\n

This connects to a broader pattern. IronCurtain enforces deterministic policy outside the model. VS Code hooks execute commands at agent lifecycle points. Anthropic\u2019s Code Review dispatches agent teams before the merge. Gemini Code Assist\u2019s Auto Approve lets the agent execute, and you review after. Each represents a different point on the agent autonomy spectrum.<\/span><\/p>\n

Plan mode sits at the conservative end \u2014 and for database migrations, major refactors, and multi-service features, that\u2019s exactly where teams want to start. Spending 20 minutes in read-only planning before the agent writes a line of code isn\u2019t overhead. It\u2019s risk management.<\/span><\/p>\n

The model routing is the quiet differentiator. High-reasoning models for planning, faster models for execution \u2014 the same strategy-vs-tactics separation that Random Labs\u2019 Slate implements at the system level, implemented here at the model routing level.<\/span><\/p>\n

The extensibility matters too. Plan mode exposes <\/span>enter_plan_mode<\/span>, <\/span>exit_plan_mode<\/span>, and <\/span>ask_user<\/span> as tools that custom extensions can build on. Teams can define organizational planning workflows and enforce policies during the research phase using plan mode as the foundation.<\/span><\/p>\n

Enter plan mode with <\/span>\/plan<\/span> in Gemini CLI. Enabled by default.<\/span><\/p>\n

Read More<\/a><\/p>\n

\u200b<\/p>","protected":false},"excerpt":{"rendered":"

The pattern across AI coding tools this week has been clear: the industry is building governance, review, and safety mechanisms […]<\/p>\n","protected":false},"author":1,"featured_media":3640,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[5],"tags":[],"class_list":["post-3639","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-devops"],"_links":{"self":[{"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/posts\/3639","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/comments?post=3639"}],"version-history":[{"count":0,"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/posts\/3639\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/media\/3640"}],"wp:attachment":[{"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/media?parent=3639"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/categories?post=3639"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/wp-json\/wp\/v2\/tags?post=3639"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}