Is Claude Code's Plan Mode Still Necessary?
Anthropic's own team questions plan mode's value in Claude Code. Here's what it still does, what's changed, and when to skip it.
What’s happening with Claude Code’s plan mode?
Plan mode in Claude Code lets you ask Claude to draft a plan before it touches any files, then approve or reject that plan before edits happen. It’s been treated as a safety step for anything complex. But the people who built it are now questioning whether it’s still needed. Boris Cherny, who created Claude Code, has said plan mode was useful and is no longer useful. A Claude Code team member named TK posted in late September that the team is considering killing the dedicated mode entirely and replacing it with effort-level controls (using shift+tab), stating plainly that he doesn’t think the models need it anymore.
That doesn’t mean planning itself is dead. It means the specific interface feature, the mode that blocks edits until you approve a plan, may be on its way out, even as the underlying practice of planning before you build stays relevant.
TL;DR
- Plan mode’s creators are walking it back: both Boris Cherny and Claude Code team member TK have said the dedicated mode is less necessary now that the models plan well on their own.
- The team is considering replacing plan mode with effort-level controls triggered by shift+tab, treating planning as a matter of reasoning depth rather than a separate UI mode.
- The strongest argument for keeping it has nothing to do with the model: one answer TK himself agreed with is that plan mode isn’t for the agent, it’s for the human reviewing the plan before changes land.
- The one concrete thing plan mode still does is block edits until you approve, which matters if you want a hard checkpoint before Claude starts editing files.
- Anthropic’s own current advice is to skip the formal mode for small jobs and only use a planning step, in or out of the dedicated mode, when a task is large, unclear, or spans many files.
- For bigger jobs, the recommended pattern is to interview Claude, write the plan to a file, then start a fresh session to execute it, rather than planning and building in the same continuous thread.
- This fits a broader pattern across Claude Code’s last ten months: skills, model switching, and context management have all shifted toward doing less manual scaffolding because the models handle more of it by default.
Why did Anthropic’s own team turn on plan mode?
The core reason is that newer Claude models reason through multi-step tasks well enough that a separate, gated planning phase adds less value than it used to. Early on, plan mode served as a guardrail against a model charging ahead and making a mess across a codebase. As the underlying models got better at holding a coherent approach across a task, the gap between “plan first, then execute” and “just execute carefully” narrowed.
TK’s public comments on this were direct: he said he mostly plans inside artifacts now, rather than using the dedicated plan mode, and he asked people who still rely on plan mode to explain why. The best answer he received, and the one he agreed with, reframed the entire debate: plan mode was never really about making the agent perform better. It was about giving the human a checkpoint. Changing a written plan costs you a few minutes. Reviewing and undoing changes spread across dozens of files costs you a lot more. That asymmetry is the real argument for planning, independent of whether the model technically needs the scaffolding to do good work.
What does plan mode still actually do?
Stripped down, plan mode’s remaining concrete function is simple: it blocks Claude from making edits until you’ve approved the plan. That’s it. It’s a gate, not a reasoning aid. If you want that hard stop before any file changes happen, the mode still provides it, for now.
Everything else people associate with “planning,” like deciding what a finished task looks like, setting the rules Claude should follow, and identifying the guardrails for a given job, isn’t exclusive to the dedicated mode. You can get the same outcome by asking Claude to interview you about requirements, writing a plan to a file, reviewing it yourself, and then opening a new session to execute. That’s the workflow Anthropic itself now recommends for large or ambiguous jobs, and it doesn’t require the formal plan mode toggle at all.
This distinction matters because it separates two things that got bundled together: the practice of planning (which remains useful) and the mode that enforces an approval gate before edits (which may not survive as a dedicated feature).
When should you still use plan mode?
Use it, or some version of explicit planning, when a task is big, unclear, or touches a lot of files. In those cases, the cost of a misunderstood scope is high, and a written plan gives you a cheap place to catch mistakes before Claude starts editing. If you’re not deeply technical, the plan also doubles as your discovery process: it’s often the fastest way to figure out what guardrails you need to specify in your prompt, because you can see where the model’s proposed approach gets things wrong or makes assumptions you didn’t intend.
Remy doesn't build the plumbing. It inherits it.
Other agents wire up auth, databases, models, and integrations from scratch every time you ask them to build something.
Remy ships with all of it from MindStudio — so every cycle goes into the app you actually want.
The recommended pattern for these larger jobs is to treat planning as a separate phase from execution entirely. Interview Claude about the task, have it write a specific plan to a file, review that file yourself, then start a fresh session to actually build it. This keeps the planning conversation from cluttering the context window that Claude uses during execution, which matters because longer context windows have been shown to reduce how reliably a model retrieves specific details buried inside them.
When should you skip it entirely?
Skip the formal planning process when you can describe the change you want in a short paragraph, or when the job isn’t a multi-day build. For small, well-scoped tasks, a quick set of goals, a definition of done, and a short list of guardrails does the same job as a formal plan without the overhead of a separate approval step. Anthropic’s own current guidance draws this line directly: small or clearly-scoped work doesn’t need the ceremony, large or ambiguous work does.
This mirrors a broader shift in how Claude Code is meant to be used. Where detailed, step-by-step scaffolding (whether that’s a plan, a skill, or a set of manual settings) used to compensate for model limitations, the newer models increasingly handle that scaffolding internally. The practical move is to try the task without the extra structure first, see where Claude actually falls short, and only add back the specific corrections or guardrails that address those gaps.
Is plan mode worth keeping in your workflow?
For most day-to-day tasks, no, not as a default habit. If you’re making a change you can describe in a sentence or two, turning on a dedicated planning mode adds a step without adding much safety, since the risk of a bad outcome is already low. For anything large, ambiguous, or spread across many files, some form of planning is still worth doing, but it doesn’t have to live inside the specific plan mode toggle. Writing a plan to a file, reviewing it, and starting fresh for execution gets you the same benefit without depending on a feature that its own creators are debating whether to keep.
The practical takeaway is to stop treating plan mode as a fixed habit and start treating planning as a judgment call based on task size and ambiguity. The mode itself might not be around in its current form much longer. The underlying discipline of deciding what done looks like before you start building isn’t going anywhere.
Frequently Asked Questions
What is plan mode in Claude Code?
Plan mode is a Claude Code feature that has Claude draft a plan for a task and blocks it from making file edits until you review and approve that plan. It’s meant as a checkpoint before changes happen.
Why do Anthropic’s own team members say plan mode is less useful now?
Both Boris Cherny, who created Claude Code, and team member TK have said that newer Claude models reason through tasks well enough that the dedicated planning step adds less value than it used to. TK has said he now mostly plans inside artifacts rather than using the formal mode, and the team has discussed replacing it with effort-level controls.
Does this mean I should stop planning before big Claude Code tasks?
Built like a system. Not vibe-coded.
Remy manages the project — every layer architected, not stitched together at the last second.
No. The argument is specifically about the dedicated plan mode feature, not about planning as a practice. For large, unclear, or multi-file jobs, Anthropic’s own recommendation is still to have Claude interview you, write a plan to a file, and start a fresh session to execute it.
What’s the actual difference between plan mode and just asking Claude to write a plan?
Plan mode enforces a hard block on edits until you approve. Asking Claude to write a plan to a file without using the mode gives you the same planning content and review opportunity, but without the enforced gate, so you’re relying on your own discipline to review before telling Claude to proceed.
When is it safe to skip planning altogether?
When you can describe the change in a short paragraph and it’s not a multi-day build. For small, well-scoped tasks, a brief list of goals, a definition of done, and any guardrails is usually enough.
