Introduction
The AI industry's competitive focus has shifted from "which model is more powerful" to "which Agent can get more work done." This guide ranks mainstream Agent tools based on real-world usage and community feedback to help beginners make an informed choice.
5th Place: TRAE
TRAE is ByteDance's long-standing Agent tool, often mentioned in discussions about domestic AI tools but rarely the first choice. It evolved from a Cursor-like intelligent IDE into TRAE SOLO, a general-purpose desktop Agent with cross-platform sync across mobile, web, and desktop.
Pros:
- Beginner-friendly, no complex configuration required
- Three-platform sync (mobile, web, desktop)
- Free for personal use with no usage limits (though peak hours may require waiting)
Cons:
- Harness engineering capabilities are weaker than mainstream competitors
- Struggles with long multi-step tasks, cross-file edits, and repeated debugging, leading to detours and high rework costs
Best For: Beginners who want to try AI Agents or have strong mobile office needs. Not recommended for heavy-duty work.
4th Place: Hermes Agent
A surprise hit after the OpenClaw craze, Hermes Agent gained attention for its long-term memory and self-evolution capabilities, sparking debates between "raising shrimp (OpenClaw)" and "raising horses (Hermes)."
Pros:
- Long-term memory that allows the Agent to learn from past experiences
- Self-evolution through autonomous skill creation
Cons:
- Over-reliance on skills can lead to outdated or incorrect workflows
- Skill selection errors can derail tasks
- Less competitive in terms of ease of use, security, and localization compared to domestic alternatives
Best For: Use as a daily assistant. Not recommended for heavy coding, PPT creation, data analysis, or complex content production.
3rd Place: OpenClaw and Its Domestic Variants
OpenClaw and its domestic derivatives (often called "domestic shrimp") are ideal for beginners due to their strong ecosystem and extensive optimizations.
Pros:
- Best domestic optimization for Chinese users
- Rich ecosystem with free usage quotas
- Easy to get started: no environment setup required, install and use immediately
Cons:
- Relies heavily on plugins and skills
- Harness capabilities are relatively weak, so performance in heavy tasks depends on plugin/skill selection
Best For: Beginners to get started with AI Agents. Domestic variants allow quick onboarding to experience Agent value, while advanced users can explore plugins, skills, and workflow configurations.
2nd Place: Claude Code
Claude Code is a staple in the Agent space, known for its exceptional harness engineering capabilities. It was the default recommendation for AI bloggers before Codex rose to prominence.
Pros:
- Strongest harness engineering in the industry
- Excellent adaptation to domestic models via CC Switch
- Best coding Agent capabilities
Cons:
- High usage barriers: CLI version requires terminal operations, which is not beginner-friendly
- Desktop version lacks Chinese language support and has complex model integration
Best For: Professional developers. For non-developers needing PPT creation, document writing, or general workflow integration, Claude Code may not be the most comfortable option.
1st Place: Codex
Codex is the new favorite in the AI space, often called the "iPhone of Agents." While its raw capabilities may not surpass Claude Code, its user experience is unparalleled.
Pros:
- Seamless integration with GPT-5.5, the most powerful general reasoning model
- Excellent plugin ecosystem with a curated marketplace (similar to the App Store)
- Best user experience with "approve for me" mode for smooth, secure operation
Cons:
- Limited usage quotas for Plus tier
- Poor adaptation to third-party models
Best For: Users with the right conditions who want a one-stop solution. Codex's strength lies in the seamless collaboration between model and execution system, making it ideal for users who prioritize experience over raw power.
Final Recommendations
- Beginners on a budget: Start with domestic OpenClaw variants. Begin with simple tasks and gradually build your AI workflow.
- Professional developers: Choose Claude Code. Even with domestic model integration, it remains excellent for development work.
- Users with the right conditions: Opt for Codex for a one-stop, seamless experience.
Final Note: Don't treat any tool as a religion. The best tool is the one that helps you get work done reliably. Focus on building the workflow that works best for you, not on chasing the most famous tool.
常见问题
I'm a total beginner — which Agent should I start with?
Start with domestic OpenClaw variants. They require zero setup — install and go. The Chinese-language optimization means you won't fight the UI while learning AI concepts. Begin with simple tasks like "summarize this document" or "help me draft an email." Once comfortable, graduate to more complex workflows. TRAE is another good free option if you need mobile sync. Skip Claude Code for now — its CLI interface will frustrate beginners before they see any value.
Why is Codex ranked #1 over Claude Code when Claude Code has stronger harness engineering?
This ranking is for beginners, not power users. Codex wins on user experience: it's a polished desktop app with a plugin marketplace, "approve for me" mode, and seamless GPT-5.5 integration. Claude Code's harness engineering is technically superior, but that advantage only matters when you're building complex multi-agent workflows — which beginners aren't doing. Think of it like cameras: a professional DSLR (Claude Code) takes better photos in expert hands, but most people get better results from an iPhone (Codex) because it handles the complexity for you.
Are all these tools free?
TRAE and domestic OpenClaw variants are free for personal use. Claude Code has a free tier with API key usage (you pay for the model's API calls). Codex requires a subscription (Plus at $20/month recommended). Hermes Agent pricing varies by region. For beginners, start with the free options (OpenClaw variants or TRAE) to learn the Agent paradigm. Only pay for Codex or Claude Code when you've outgrown what the free tools can do.
Can I use multiple Agents together?
Yes, and many professionals do. A common pattern: use Claude Code for heavy coding tasks, Codex for general productivity and PPT creation, and an OpenClaw variant for quick daily tasks. The tools don't compete — they complement each other. The key is building a workflow where each tool handles what it does best, rather than trying to make one tool do everything.