Tool of choice - How to code with LLM #1

Let's start this series by opting for our tools of choice.

Since there are many, let's break down the most popular ones by pros and cons, and identify which might be more suitable for your existing workflow or subscriptions.

IDE or Terminal?

The first distinction we make is whether you want to replace or enhance your existing IDEs, or prefer an agnostic approach with a CLI.
If you are looking to enhance your existing IDE workflow and you are already part of the JetBrains product family, the simplest way is to leverage Junie.

Since JetBrains products can be pretty heavy and you might be on the lighter side of things, there are several “forks” that you can switch to with the ability to import config from your existing VSCode setup, like Cursor or Windsurf. Other alternatives for VSCode users who do not want to replace their existing setups include plugins such as Cline or Kilo.
For you who might feel more comfortable in your Terminal or do not want to switch or entangle your AI coding with your editor, there are many state-of-the-art options for you, including ClaudeCode, OpenCode, CodexCli, and more.

Now that we’ve got the most significant difference in the buffet of AI coding tools out of the way, let's go through the major ones and compare their strengths and weaknesses. I’ll be skipping ones like Windsurf, Cline, or Kilo, since they are often very similar and often compete mainly on pricing.

Cursor

This is one of the biggest names in the space, and it’s excellent in easing you into agentic coding, since it provides powerful autocompletion with excellent agentic capabilities, resulting in a git diff-like view of LLM proposed changes to files.

Pros:

• Simple to set up, very much a battery-included solution
• Friendly GUI
• Excellent for transitioning from coding to agentic coding
• Allows you to bring your own models
• Built-in web browser that LLM can interact with
• If you are using VSCode, you will be right at home

Cons:

• Proprietary
• At the mercy of Cursor pricing for subscription plans
• Quite costly for fully agentic development
• No transparency in model automatic selection (mode for saving cost)
• Even when bringing your own models, you still need at least a $20 subscription
• If you dislike VSCode, you will be right in hell
  

Claude Code

A leading CLI tool that most people have probably heard of. Most innovative tool introducing features such as mcp, skills, and sub-agents. Made by Anthropic, the company behind the Opus and Sonnet models, and the Model Context Protocol itself, the quality of ClaudeCode and its subscription model, which grants generous usage of its models, has proven to be an unstoppable force.

Even if you are not a fan of CLI tools, Claude Code has you covered with things like official plugins for JetBrains.

Pros:

• Access to state-of-the-art models with a subscription
• Generous usage in comparison to the price of the subscription
• Polished terminal experience backed by a large company pushing the envelope
• A huge community and ecosystem of community-made plugins
• Independent of IDE, while having official plugins for tighter integration

Cons:

• Anthropic models are generally expensive; for “power users,” expect a subscription of at least $100–200
• Silently removing the ability to use other models in ClaudeCode
• Anthropic is actively fighting against the ability to use their ClaudeCode subscription in other CLI tools
  

Codex CLI

Where Claude Code is the staple of Agentic CLI tools, Codex is the newest kid on the block. While their features are lacking, they also offer good-enough IDE plugins if you are not a fan of the terminal. Why even talk about this? There’s a high chance you are familiar with ChatGPT, and if you are subscribed to ChatGPT Plus, you already have access to Codex CLI.

Pros:

• Comes with ChatGPT Plus
• Seems to have higher usage for the equivalent of the ClaudeCode Pro plan
• Simple, yet effective plugins for the IDE
• Possible to interact with OpenAI models through subscription in 3rd party tools like OpenCode

Cons:

• Limited to the OpenAI models (even though they are strong)
• Even though 20$ plan will give you plenty of usage, taking it a step further will put you at 200$ (10x price, 5x usage)
• Terminal experience is subpar
• Yaml configuration, while the rest of the competition is using similar JSON configs
• Strange scoping between the global and project configuration
  

OpenCode

Fully open source, the best terminal experience you can ask for. Since it’s not a product of any company creating models, it's completely model-agnostic. This is the place where you can integrate many different providers and test out models, allowing you to choose precisely the right models for the given task and those that you feel “work for you well”. Even if you do not have a subscription yet, OpenCode offers free model sets to try (no registration required).

If you are not a fan of Terminal, they also offer a web client.

Pros:

• Model agnostic
• No vendor lock-in
• The best terminal experience of all CLI tools
• Huge community
• Allow the use of less-known models and their code plans, which are often close to state-of-the-art models, for a fraction of the price

Cons:

• Stability and level of polish. It can be quite unstable on Windows
• Choice overload when selecting models
• If you are planning to use a ClaudeCode subscription with OpenCode, prepare for neverending battle
  

As I have already said, there are many tools I haven’t even mentioned, and this pros-and-cons list may only be relevant for about a week. You might already have your favourite. You might still be unsure which one to go for, since all of them sound good. It doesn’t matter in the end; each of them does the job. Remember, it’s still very early. Do not get hung up on one thing; stay curious, explore, and experiment.