Dual Mode System
Caret's dual mode system presents a completely different development philosophy from Cline. While fully preserving Cline's Plan/Act mode, it additionally provides two new ways of working. This allows developers to choose how they collaborate with AI based on the situation, creating an optimal environment tailored to the nature of their work and personal preferences.
Understanding the Philosophical Differences: Cline vs Caret
Cline and Caret each have distinct development philosophies, which become clearly evident in how they facilitate interaction with AI. While Cline pursues structured development through strict separation between planning and execution phases, Caret is designed with an emphasis on developer flexibility and control.
Cline's Plan/Act Philosophy
Cline faithfully follows the traditional software development principle of "first plan, then execute." This system operates in two clearly distinct modes.
In Plan mode (planning phase), Cline carefully plans the work. During this phase, it analyzes the codebase, understands requirements, and formulates a step-by-step execution strategy. It waits for developer approval and modifies no files. Like an architect drawing blueprints, it focuses entirely on planning everything first.
When switching to Act mode (execution phase), Cline puts the plan into action. However, there's one important characteristic here. Once execution begins, Cline continues executing until the developer explicitly stops it. If you ask "wait, how does this work?" during execution, Cline focuses on completing the task rather than engaging in conversation.
This approach is highly effective for structured development, but it has one limitation: when questions arise or mid-task verification is needed during execution, it's difficult to respond flexibly.
Caret's Agent/Chatbot Philosophy
Caret respects Cline's philosophy while proposing two completely different ways of working. Caret's core principles are flexibility and developer control.
Agent mode views AI as an autonomous partner. The biggest feature is the ability to immediately switch to conversation even during execution. If you ask "wait, what exactly is JWT?" while code is being written, the Agent immediately switches to explanation mode. After the explanation, simply say "got it, continue" and work resumes. This provides an experience very similar to actual pair programming with a human.
Chatbot mode has the opposite philosophy. Here, AI focuses on an advisor role. The developer codes directly, and AI only provides advice when needed. When you think "I feel like I'm relying too much on AI," "I want to maintain my skills," or "I want to learn new technologies by actually using them," Chatbot mode shines.
Detailed Comparison of 4 Modes
Understanding Cline and Caret's 4 modes makes it clear which mode is suitable for each situation.
| Characteristic | Cline Plan | Cline Act | Caret Agent | Caret Chatbot |
|---|---|---|---|---|
| Core Philosophy | Planning phase | Execution phase | AI partnership | Developer-led |
| Conversation Flexibility | ✅ Always possible | ❌ Executes until stopped | ✅ Instant switching | ✅ Full control |
| Autonomy Level | 🔒 Awaits approval | 🚀 Auto-executes | 🤝 Collaborative | 👨💻 Passive |
| Execution Method | Plans only | Continues until stopped | Flexible execution | Provides advice |
| Developer Role | Approver | Observer | Collaborator | Leader |
| Learning Effect | Moderate | Low | Moderate | ⭐ Very High |
Caret Agent Mode: Flexible AI Partner
The essence of Agent mode is "the flexibility to converse whenever needed." This is also the biggest difference from Cline's Act mode.
How It Works
In Agent mode, Caret works autonomously while remaining constantly open to developer intervention. For example:
Developer: "Implement user authentication with OAuth2"
Agent: [Writing token management logic...]
Developer: "Wait, where is it safe to store the refresh token?"
Agent: "Great question. From a security perspective, refresh tokens should..."
[Provides detailed explanation]
Developer: "Got it, let's use httpOnly cookies. Continue"
Agent: [Resumes writing code with httpOnly cookie storage logic...]
This way, Agent mode maintains workflow while instantly transitioning to educational conversation when needed.
When Should You Use Agent Mode?
Agent mode is particularly effective in the following situations:
When implementing complex features: Suitable for work requiring modification of multiple files with intermittent verification needs. For example, when implementing a payment system and questions arise about PG integration methods, you can immediately ask, receive an answer, and continue work.
When learning while developing: Useful when you want to learn about new technologies or libraries while using them. If AI is writing code and you ask "why are you doing it this way?", you'll get an immediate answer.
When quickly creating prototypes: Agent mode is perfect when you want to realize ideas quickly but adjust direction along the way.
Caret Chatbot Mode: Developer-Led Development
Chatbot mode has a clear philosophy of "AI advises only, I code." This mode is chosen when you can't fully trust AI or want to maintain your own development skills.
How It Works
In Chatbot mode, rather than writing code directly, Caret provides guidance so developers can write it themselves:
Developer: "Show me how to manage server state with React Query"
Chatbot: "Server state management using React Query proceeds in these steps.
Step 1: QueryClient setup
In App.tsx, you need to wrap your app with QueryClientProvider.
[Explanation of setup example]
Step 2: Using useQuery hook
When fetching data..."
Developer: [Reads explanation and writes code directly]
Developer: "I wrote it like this, how do I set cache time?"
Chatbot: "Use staleTime and cacheTime options..."
Developer: [Modifies directly again]
When Should You Use Chatbot Mode?
Chatbot mode is most valuable in the following situations:
When studying new technologies: When you want to learn by writing directly rather than having AI simply write code, Chatbot mode is optimal. When learning a new framework in a study group or coding bootcamp, using this mode maximizes learning effects by receiving AI advice while actually writing code by hand.
When you can't fully trust AI: If you think "I'm worried AI might mess up my codebase," it's safer to receive AI suggestions in Chatbot mode and write code yourself.
When you want to maintain development skills: With frequent AI tool use, concerns may arise like "isn't my ability to code directly deteriorating?" Chatbot mode addresses these concerns while still leveraging AI knowledge.
When making important architecture decisions: When making major technical choices like microservices vs monolithic, SQL vs NoSQL, you can discuss the pros and cons of each option in depth with Chatbot, with the developer making the final decision.
Cline Compatibility: Consideration for Existing Users
If you're a Cline user encountering Caret for the first time, don't worry. Caret fully preserves Cline's Plan/Act mode exactly as it is.
When you select Cline mode in settings, Caret operates exactly identically to original Cline. You can use the familiar Plan → Act workflow as is, with all shortcuts and behaviors remaining the same. This is one of Caret's core philosophies: "Provide new choices while ensuring backward compatibility."
Cline user → Install Caret → Select Cline mode → Completely identical experience as before
Whenever you want to try Caret's new modes, simply switch to Agent or Chatbot mode in settings. And returning to Cline mode is equally free.
Mode Selection Guide: Optimal Mode for Each Situation
If you're struggling with which mode to choose, refer to this guide.
When Agent Mode Is Suitable
In situations where you can trust AI, Agent mode becomes a powerful productivity tool. When developing typical CRUD applications, performing repetitive refactoring work, or working with familiar technology stacks, it's efficient to delegate much to AI and intervene only when necessary.
Agent mode is also useful when rapid prototype development is needed. When a startup needs to quickly build an MVP or urgently prepare a demo to validate an idea, Agent mode's autonomy provides significant help.
When Chatbot Mode Is Suitable
When learning is the main purpose, Chatbot mode is strongly recommended. If you're learning a new programming language, becoming familiar with an unfamiliar framework, or deepening understanding while solving algorithm problems, writing code directly is much more effective.
Chatbot mode is also advantageous for projects where code quality is extremely important. When developing systems in fields like finance, healthcare, or aviation, or production services used by millions of users, it's safer to directly review and write every line.
Chatbot mode also shines in team study or pair programming situations. The process of receiving AI advice while discussing with learning colleagues and directly writing code maximizes learning effects.
When Cline Mode Is Suitable
If you have existing Cline usage experience and that workflow is ingrained, continuing with Cline mode is an excellent choice. If learning a new approach feels burdensome or the clear distinction between planning→execution phases matches your work style, there's no need to change.
Cline mode is also suitable for developers who prefer explicit approval steps. If you feel that thoroughly reviewing in Plan mode and executing intensively in Act mode is more systematic, maintain Cline mode.
Real-World Application Scenarios
How can these modes be utilized in actual projects?
Scenario 1: Startup MVP Development
Situation: Must create an MVP for investor demo within 3 weeks.
Weeks 1-2: Agent mode
→ Quickly implement basic features
→ Ask questions immediately as they arise
Week 3: Switch to Chatbot mode
→ Directly review and rewrite core logic
→ Directly write security-related code with AI advice
Day before demo: Switch back to Agent mode
→ Quickly handle UI polishing and bug fixes
Scenario 2: Learning New Technology Stack
Situation: Project requiring transition from React to Vue.js
Phase 1: Chatbot mode (1-2 weeks)
→ Learn Vue.js basics
→ Write simple components directly while learning
→ Receive code reviews from AI
Phase 2: Switch to Agent mode
→ Quickly implement features with familiar patterns
→ Switch back to Chatbot only for complex parts to learn
Scenario 3: Legacy Code Refactoring
Situation: Modernizing 5-year-old legacy codebase
Phase 1: Chatbot mode
→ Request existing code analysis
→ Discuss refactoring strategy
→ Make core architecture decisions directly
Phase 2: Agent mode
→ Perform repetitive refactoring according to defined patterns
→ Automatically convert hundreds of files
Phase 3: Return to Chatbot mode
→ Final review and directly handle exception cases
How to Switch Modes
Switching between modes is very simple.
Switching in Settings:
- Open Caret settings window (⚙️ icon)
- Navigate to the General tab
- Select desired mode in AI Mode section:
- Cline Mode (Plan/Act)
- Caret Agent Mode
- Caret Chatbot Mode
- Changes apply immediately
Project-specific settings are also possible. Create a .vscode/settings.json file in your project root:
{
"caret.aiMode": "agent" // or "chatbot", "cline"
}
With this setting, the specified mode automatically activates whenever you open that project.
Frequently Asked Questions
Q: Is Agent mode better than Cline's Act mode?
A: It's difficult to say one is "better." The two modes have different philosophies. Cline Act is optimized for completing work without interruption, continuing execution until stopped after approval. Caret Agent can flexibly transition to conversation even during execution, making it more suitable for learning while developing or when mid-task verification is needed. Choose based on situation and personal preference.
Q: Does Chatbot mode never write code at all?
A: No. Chatbot mode can also generate code when requested. However, it's fundamentally focused on providing explanations and guidance so developers can write directly. If you request "show me this part in code," you can always receive code examples to reference for direct writing or use as-is.
Q: Will problems occur if I switch modes frequently?
A: No problems at all. Each mode operates completely independently, but conversation history and context are shared across all modes. Even if you work in Agent mode and switch to Chatbot mode, Caret remembers all previous conversation content. Work is never reset and no information is lost, so confidently switch modes as situations require.
Q: Is it okay to change modes mid-project?
A: Absolutely. Changing modes according to project phases can actually be effective. For example, quickly build an initial prototype in Agent mode, carefully develop core features in Chatbot mode, and handle final polishing in Agent mode again. Use flexibly.
Next Steps
If you understand Caret's dual mode system, proceed to the next steps:
- Customize AI with Persona System - Adjust AI's tone and personality to your preferences.
- Learn About Rule Priority System - Control AI behavior with project-specific rules.
- Start Free with Caret Provider - Learn how to start immediately without an API key.
The dual mode system is exclusive to Caret. While fully preserving Cline's Plan/Act mode, it additionally provides two new choices: Agent and Chatbot. Each mode has unique philosophy and advantages, and developers can freely choose the most suitable mode according to their situation.