In a recent LinkedIn post, Kobi Omenaka discusses critical safety features within Claude Code, warning users about potential pitfalls that can lead to accidental data loss. Omenaka, who is documenting his “Claude Code 101” series to help users extract practical value from the AI tool, highlights that many beginners inadvertently use the most dangerous operational mode.
The core of Omenaka’s warning revolves around understanding the different “gears” or modes of operation within Claude Code. He emphasizes that lack of caution and planning is the primary reason users might find themselves exclaiming, “It deleted my whole project.”
“If you use Claude Code long enough, you will say that one day. Usually the day you let it run with no plan and no brakes.”
Omenaka breaks down Claude Code’s operational modes into three distinct “gears,” with the most perilous one often being engaged unintentionally.
Understanding Claude Code’s Operational Modes
According to Kobi Omenaka, Claude Code operates with three primary modes, each with different implications for user control and speed. He categorizes these as follows:
1. Plan Mode
This is the mode Omenaka strongly advocates for users to adopt. It is activated by pressing “Shift and Tab twice.” In this mode, Claude Code first reviews the user’s request, thinks through the implications, and presents a plan for review. No changes are made to the project until the user explicitly approves the plan. Omenaka likens this to a builder presenting architectural drawings before construction begins.
2. Ask First Mode
This is the default setting for Claude Code. While slower than other modes, it ensures that the AI seeks user confirmation before executing any edits. This provides a crucial layer of oversight, preventing unintended modifications.
3. YOLO Mode
Formally known as “Bypass Permissions,” this mode allows Claude Code to operate at maximum speed by executing tasks without seeking explicit user approval for each step. While this can significantly accelerate development, Omenaka warns it is the mode most associated with accidental data erasure. As Kobi Omenaka puts it, “no questions is exactly how people wipe their own work.”
The Importance of Planning in AI Development
Omenaka argues that the exceptional results achieved by some users are not due to access to a superior AI model, but rather their disciplined approach to planning. He stresses the importance of proactive planning, stating:
“The people who get incredible results are not on some secret model. They just plan first.”
His advice is to consistently engage “Plan Mode” by using the “Shift, tab, tab” shortcut, reviewing the proposed plan, making necessary adjustments, and only then allowing the AI to proceed. This methodical approach, he suggests, is analogous to responsible project management in any field.
“Strong foundations up front means far less firefighting later.”
As Omenaka continues his “Claude Code 101” series, he plans to delve deeper into briefing the AI effectively, likening it to briefing a contractor to ensure the initial plan generated is accurate and requires minimal revision. This focus on robust initial planning, he contends, is key to leveraging AI tools like Claude Code safely and productively, avoiding the costly mistakes that can arise from unchecked AI execution.
📝 About This Content
This article is based on insights shared by Kobi Omenaka on LinkedIn.
📅 Originally posted on June 18, 2026 | View original post on LinkedIn →