# No, Gemini CLI CAN’T touch Claude Code BUT I’d NEVER bet against Google

Source: https://www.youtube.com/watch?v=Go9W_h7kcjE
Recap page: https://rapidrecap.app/video/Go9W_h7kcjE
Generated: 2025-07-17T09:04:39.396+00:00

---
## Quick Overview

Google's newly released Gemini CLI, while incredibly fast and capable, currently falls short of Anthropic's Claude Code in terms of precision, comprehensive feature implementation, and overall agentic intelligence in UI component generation. However, Gemini CLI's powerful underlying models, programmatic execution capabilities, and Google's vast resources position it as a strong future contender that should not be underestimated.

**Key Points:**
- Gemini CLI, powered by Gemini 2.5 Pro, demonstrates superior speed in basic tasks like dependency installation and initial UI component generation compared to Claude Code (Opus 4).
- Claude Code consistently excels in precision and comprehensive feature implementation for complex UI components, often producing cleaner and more functional results than Gemini CLI.
- Gemini CLI features an automatic memory system that learns user preferences (e.g., not to auto-run dev servers) and can be prompted to remember specific instructions, a unique capability.
- Claude Code's structured planning and sub-agent support allow it to manage multi-component tasks more effectively, breaking down complex problems into manageable steps.
- Gemini CLI's tendency to delete previous components by default (though fixable with explicit memory prompts) and its less refined UI styling in some instances are notable drawbacks.
- Despite current performance gaps in certain complex tasks, Gemini CLI's powerful underlying models, robust documentation, and Google's vast resources position it as a strong future contender in the agentic coding space.
- The video emphasizes that while Claude Code is currently the most 'cracked' tool for engineering, developers should remain dynamic and not 'bet against Google' given their potential for rapid innovation.

![Screenshot at 0:03: Google blog post announcing Gemini CLI, an open-source AI agent.](https://ss.rapidrecap.app/screens/Go9W_h7kcjE/00-00-03.png)

**Context:** The video delves into the competitive landscape of AI coding tools, specifically comparing Google's recently launched Gemini CLI with Anthropic's established Claude Code. It highlights the broader trend of major tech companies, including OpenAI, developing command-line interface (CLI) tools that leverage large language models (LLMs) to assist developers directly within their terminals. The video sets the stage by acknowledging the historical context of innovation and 'copying' in the tech industry, tracing it back to foundational AI research like Google's 'Attention Is All You Need' paper. The core objective is to evaluate whether Gemini CLI offers genuinely new capabilities that surpass Claude Code, the current leader in agentic coding, or if it primarily serves as Google's entry into an already burgeoning market.

## Detailed Analysis

The video conducts a hands-on comparison between Google's new Gemini CLI (running Gemini 2.5 Pro) and Anthropic's Claude Code (running Claude 4 Opus) for agentic coding tasks, specifically focusing on UI component generation. The comparison uses a custom 'Agentic Coding Tool Eval' codebase with five levels of UI complexity. Initial setup involves creating isolated Git worktrees for each tool and configuring their respective AI models. In early tests, Gemini CLI demonstrates superior speed in installing dependencies and generating basic components, often completing tasks before Claude Code finishes planning. A unique feature of Gemini CLI is its automatic memory saving, allowing it to remember user preferences (e.g., not to auto-run dev servers). However, Gemini CLI initially struggles with retaining previously generated components, often deleting them by default, a behavior that requires explicit prompting to correct. As complexity increases, Claude Code consistently proves more precise and thorough, implementing features like icons in stats cards and cleaner carousel designs, while Gemini CLI sometimes produces UI cruft or misses details. Claude Code's structured planning and step-by-step execution, akin to an engineer's workflow, allow it to manage complex tasks without losing context or deleting prior work. In contrast, Gemini CLI's rapid, less structured approach, while fast, can lead to errors or incomplete implementations in higher-level tasks like multi-step forms and Kanban boards, where Claude Code excels in functionality and visual fidelity. Despite these current shortcomings, Gemini CLI offers powerful features like direct access to Gemini models, non-interactive mode for scripting, robust documentation, and automatic memory. The video emphasizes that while Claude Code is currently superior for engineering tasks, Google's immense resources, continuous model development, and ownership of vast data sources like YouTube make it a formidable long-term competitor in the agentic coding space, urging developers not to 'bet against Google'.

### Introduction to Agentic Coding Tools

- Google's Gemini CLI is introduced as a new open-source AI agent, positioned alongside OpenAI's Codex CLI and Anthropic's Claude Code, which is dubbed the 'King of Agentic Coding'
- The tech ecosystem is noted for 'rip-offs' and lack of credit, a trend traced back to Google's 'Attention Is All You Need' paper
- The core question is whether Gemini CLI offers net new capabilities that outperform Claude Code.

### Comparison Setup and Initial Tests

- A custom 'Agentic Coding Tool Eval' codebase is used for side-by-side comparison, with agents operating in isolated Git worktrees
- Claude Code runs on Opus 4, while Gemini CLI runs on Gemini 2.5 Pro
- Initial tasks like dependency installation and HTML title updates show Gemini CLI is faster, but Claude Code excels in planning and structured execution.

### Level 1-3 UI Component Evaluation

- For basic components (Hero Banner, Stats Card), Gemini CLI is faster but sometimes deletes previous content, requiring explicit memory prompts
- Claude Code's Hero Banner is cleaner, and its Stats Card correctly implements icons where Gemini CLI uses generic dots
- For interactive components (Buttons), Gemini CLI's speed is 'insane', but it continues to delete prior work until explicitly instructed to remember, a unique automatic memory feature.

### Level 3-5 Intermediate and Expert Components

- Claude Code consistently demonstrates superior precision and completeness in complex tasks like Card Carousels and Data Tables, producing cleaner UI and more robust functionality (e.g., proper image handling, sorting, pagination)
- Gemini CLI, despite its speed, often requires re-prompting or produces UI cruft in these complex scenarios
- For advanced components like Multi-Step Forms and Kanban Boards, Claude Code's detailed planning and step-by-step approach lead to fully functional and visually complete implementations, while Gemini CLI struggles with integration and specific features like drag-and-drop.

### Gemini CLI's Strengths and Weaknesses

- Pros include direct access to powerful Gemini models, programmatic execution via non-interactive mode, good documentation, and an automatic memory system
- Cons include lack of custom slash commands, no sub-agent support (cannot parallelize tasks like Claude Code), less detailed debugging traces, and a tendency to delete previous work by default (though fixable with memory prompts)
- The video notes a 'lack of innovation in product culture' as Gemini CLI often appears to be a 'rip-off' of existing tools.

### Overall Conclusion and Future Outlook

- Claude Code is currently deemed the superior agentic coding tool due to its precision, planning capabilities, and comprehensive implementations
- However, Google's vast resources, continuous model development, and ownership of massive training data (like YouTube) make it a formidable long-term contender
- The video advises against 'betting against Google' and encourages developers to focus on the unique capabilities tools offer rather than brand loyalties, emphasizing the importance of the terminal as the highest-leverage tool for engineers.

![Screenshot at 0:03: Google blog post announcing Gemini CLI, an open-source AI agent.](https://ss.rapidrecap.app/screens/Go9W_h7kcjE/00-00-03.png)
![Screenshot at 0:05: OpenAI Codex CLI website, showing its interface for delegating tasks to a software engineering agent.](https://ss.rapidrecap.app/screens/Go9W_h7kcjE/00-00-05.png)
![Screenshot at 0:10: Anthropic's Claude Code website, highlighting it as 'Your code's new collaborator'.](https://ss.rapidrecap.app/screens/Go9W_h7kcjE/00-00-10.png)
![Screenshot at 0:56: Side-by-side terminal view showing Claude Code (left) and Gemini CLI (right) interfaces for agentic coding evaluation.](https://ss.rapidrecap.app/screens/Go9W_h7kcjE/00-00-56.png)
![Screenshot at 1:31: UI Component Evaluation README, outlining 5 levels of UI prompts by complexity (Basic Static, Simple Props, Intermediate, Advanced, Expert).](https://ss.rapidrecap.app/screens/Go9W_h7kcjE/00-01-31.png)
![Screenshot at 2:15: File explorer showing duplicated codebases for 'claude_code' and 'gemini_cli' within the 'trees' directory, enabling isolated testing.](https://ss.rapidrecap.app/screens/Go9W_h7kcjE/00-02-15.png)
![Screenshot at 3:50: Side-by-side terminal view showing Claude Code (left) and Gemini CLI (right) processing the 'install dependencies and update HTML title' task, with Gemini completing faster.](https://ss.rapidrecap.app/screens/Go9W_h7kcjE/00-03-50.png)
![Screenshot at 5:10: Gemini CLI's generated 'Hero Banner' UI component displayed in a browser, showing a purple-to-blue gradient background with centered text.](https://ss.rapidrecap.app/screens/Go9W_h7kcjE/00-05-10.png)
![Screenshot at 7:30: Gemini CLI's generated 'Stats Card' UI component, showing 1.2M Users, 99.9% Uptime, and 24/7 Support metrics with generic dots at the top.](https://ss.rapidrecap.app/screens/Go9W_h7kcjE/00-07-30.png)
![Screenshot at 9:27: Gemini CLI's generated 'Button Examples' UI component, showcasing various button variants (Primary, Secondary, Danger), sizes (Small, Medium, Large), and states (Loading, Disabled).](https://ss.rapidrecap.app/screens/Go9W_h7kcjE/00-09-27.png)