3 Different Methods to Leverage LLMs while Programming

This guide covers three approaches to integrating AI assistants into your development workflow, ranging from basic to advanced. Each method builds on the previous one, helping you find the right balance of automation, control, and efficiency for your projects.

1. Using Any LLM's Chat Interface

The most straightforward approach is using an LLM's chat interface directly. This method requires no additional setup—simply copy your code, paste it into the chat, and ask questions or request changes.

Setup

• Use tools like Copy4AI to easily multi-select and copy multiple files from VSCode
  • In VSCode file explorer, command + click through each directory and file, then right-click and select "Copy to Clipboard (Copy4AI)"
  • Paste directly into your LLM of choice, such as Deepseek or Claude
• Request the AI to add context to code snippets, update entire files, or explain complex logic

Best Practices

• Provide clear, specific instructions
• Include relevant context about your project setup
• Ask for explanations when you're unsure about suggested changes

This method works well for occasional assistance or when you're exploring ideas without committing to a specific tool.

2. IDE Addons: Copilot, Cursor, Roo Code

For tighter integration, IDE addons bring AI capabilities directly into your editor.

Copilot is now native to VSCode and offers quick intellisense-style suggestions for free. While Copilot and similar tools like Cursor provide excellent autocomplete features, their agentic capabilities are limited.

Roo Code (GitHub) is a VSCode extension that enables agentic workflows with any model provider. It supports MCP (Model Context Protocol) frameworks, allowing for more sophisticated tool usage and context management.

Other options include:

• Kilo Code (forked from Roo Code)
• Gemini Code Assist

Tips

• Use free model APIs (like Chute) to test different addons before committing
• Configure the extension to understand your project structure for better context

These tools bridge the gap between manual copy-paste and full automation, offering good balance for day-to-day development.

3. Claude Code and CLI-Based Agents

For complete automation and control, CLI-based AI agents represent the most powerful option. Claude Code leads this category with robust agentic capabilities and MCP support.

CLAUDE.md: Project Configuration

One of the most valuable features is the CLAUDE.md file—a configuration document placed in your repository root that defines how the AI assistant should operate. This file can specify:

• Automation rules for branch creation and PR workflows
• Code style preferences and conventions
• Testing requirements and quality gates
• Project-specific context that helps the AI understand your codebase

Example Automation Configuration

Here's a practical setup for automated workflows:

Automation

After completing a feature from the main branch:

1. Create a descriptive branch (e.g., feature/user-authentication or fix/login-bug)
2. Run tests and validation
3. Commit changes with a clear message
4. Push the branch and create a Pull Request automatically
5. Provide the PR link immediately

Do not ask for validation on branch names or PR titles—execute the workflow without waiting for user input, unless there's a blocker (merge conflicts, failing tests, etc.).

This level of automation is ideal for repositories with established patterns and CI/CD processes.

Multi-Project Workflows

Advanced users often combine Claude Code with:

• Git worktrees to manage multiple branches simultaneously
• TMUX sessions for detached terminal management
• SSH access for remote development servers
• Web interfaces to monitor multiple agent sessions

These techniques allow you to run several AI-assisted workflows in parallel, dramatically increasing throughput.

Other CLI Options

Several other tools offer similar experiences:

• Qwen Code
• Grok Code
• Groq Code

While generally cheaper, they may have fewer features or less polish compared to Claude Code.

Choosing the Right Approach

Method	Best For	Effort Required
Chat Interface	Casual use, one-off questions	Minimal
IDE Addons	Daily development, autocomplete	Low
CLI Agents	Full automation, complex workflows	Medium to High

Your choice depends on your needs:

• Start with method 1 if you're trying AI coding tools for the first time
• Move to method 2 for integrated daily use
• Adopt method 3 when you need to automate repetitive tasks and enforce consistent workflows

Final Recommendation

For teams and active projects, method 3 with a well-configured CLAUDE.md delivers the best long-term value. The initial setup pays for itself quickly through consistent automation and reduced manual overhead.

For individual developers or smaller projects, method 2 provides a good balance of convenience and capability.

Regardless of which method you choose, the key is to start simple, iterate on your process, and find what fits your workflow. AI assistance is a tool—the more thought you put into configuring it for your needs, the more effective it becomes.