Start here! Choose your tool!

25 May 2025 • 2 min read

Start by choosing the right tools for AI-powered software development.

You can pick between cloud-based platforms or tools that are directly integrated into your development environment. For the examples in the blog posts on this site, we use Cursor to demonstrate how these AI tools work in practice.

These tools generally fall into two categories: LLM-only and agent-based. Understanding the difference between them can help you choose the right assistant for your workflow.

LLM-Only vs. Agent-Based Coding Tools

LLM-only tools use a single large language model (LLM), like GPT, to respond to prompts. These tools offer code suggestions, completions, and edits in real time, usually within your existing editor. They’re reactive, responding to what you type or ask.
Agent-based tools go further. They simulate reasoning and planning, using multiple models or processes to break down tasks, maintain context, and carry out iterative actions—like writing a file, testing it, debugging, and improving it over time. They feel more like collaborators than just assistants.

Popular Tools in the AI Coding Ecosystem

1. GitHub Copilot (LLM-Only)

GitHub Copilot integrates seamlessly with IDEs like VS Code and JetBrains, offering real-time code completions and suggestions. It excels at filling in boilerplate, writing functions, and helping with syntax, but it doesn't plan or reason beyond each prompt.

2. Cursor and Windsurf (Agent-Based, integrated in IDE)

Cursor and Windsurf are two agent based coding tools integrated in the IDE. The are both forks of Developer Studio Code and look similar. Both can be used for building entire software projects through AI agents. It works by breaking user intent into high-level goals and then using tasks to handle planning, coding, and testing. It’s great for generating new apps, refactoring large systems, or maintaining legacy codebases.

3. OpenAI Codex and Manus (Agent-Based, cloud)

Both Codex and Manus are cloud-based software engineering agents that can work on many tasks in parallel. They can perform tasks for you such as writing features, answering questions about your codebase, fixing bugs, and proposing pull requests for review; each task runs in its own cloud sandbox environment, preloaded with your repository.

Final Thoughts

The line between tools that help and tools that build is blurring fast. LLM-only tools like Copilot boost your speed and productivity with smart suggestions, while agent-based tools like Cursor, Windsurf, Codex, and Manus are moving toward true AI collaboration—able to plan, reason, and adapt over time.