AI IDEs are changing drastically by the week. Always check the
latest features and pricing.
Building with LLMs is not just about writing Python scripts anymore. A Software
Engineer or Developer is juggling API keys, managing different model providers,
starting and stopping Docker containers, making sure you do not commit API keys,
and constantly iterating on complex pipelines. The right IDE becomes your
command center.
My needs:
AI which had context of my codebase
AI which made diff changes but I still need to review and approve
Close to what I was using, VSCode
At this point, AI IDEs have matured significantly. But I am always a bit
skeptical...
Looks just like VSCode, but with AI assistance. You are able to switch between
models like Sonnet, GPT-4o, Gemini, etc. along with their "thinking" variants.
Pros:
Native VSCode compatibility with all my existing extensions
The AI pair programming felt genuinely helpful, not intrusive
Excellent at understanding context across large codebases; along with random
snippets of text with "notes"
AI which made diff changes but I still need to review and approve
Easily switch between models like Sonnet, GPT-4o, Gemini, etc. along with
their "thinking" variants
Add images to help the LLM understand the context; for example, I gave the LLM
an image of a color palette and it was able to generate a Tailwind CSS class
based on the color palette
You can connect MCPs to your project
Cons:
Background Agent has the same performance as the regular Cursor agent
Cannot move the Cursor Chat window
Cursor pricing is not straight-forward and recent changes caused developer
discontent
I was shocked at how my friend who has no coding experience, was able to create
a web app with a set of prompts. The app has a simple Tailwind UI with simple
password login and user management.
Pros:
Low setup time: from idea to running code in minutes
Excellent for rapid prototyping
Hosting was free has fee plan but you need paid to make it worth it
Hosting and DevOps were an afterthought since Replit took care of hosting and
deploying
Built-in NoSQL, SQL, and Redis databases
Cons:
Code generator is proprietary and cannot choose the model
Limited control over the underlying environment
Awful for integrating with specific tools and frameworks such as using your
existing Firestore database
Though the AI debugged itself, it would sometimes get stuck and having to
debug it manually was a challenge because it would be the first time looking
at the source code
You can run on terminal so great for headless development or if you just
prefer terminal editors like nano or vim
Part of Claude's Pro plan so no need to worry about pricing
Works best if your IDE has a terminal in which case, integrates better with
your workflow
You can process images by dragging and dropping them into the
terminal
You can connect MCPs to your project
Customizable CLAUDE.md file to set up your project for programming
conventions, instructions, and how to structure your codebase
Cons:
There is no GUI so you have to use the terminal
You can have multiple terminals working on the same project but you may have
to deal with merge conflicts
My Final Decision
Previously...
I decided on Cursor.
✅ AI which had context of my codebase
✅ AI which made diff changes but I still need to review and approve
✅ Close to what I was using, VSCode
Now...
I switched from Cursor to Claude Code. As I found out more about Claude Code
features such as the CLAUDE.md file, ability to pull Github Issues and push
Github PRs, along with fact Claude Code is part of Claude's Pro plan which I
already had, the choice was more clear.
I could still use Cursor or VSCode as an editor but now I can have Claude Code
open in the in-editor terminal.
Price
Cursor has always had friction and controversy with its pricing model.
Personally, I was not sure when I would hit the token-based limit for my $20 /
month tier plan. In the last month I did hit the limit and started paying at a
variable rate.
With Claude Code, the pricing is more straightforward. You pay $20 / month for
the Pro plan and $20 / month which includes all the features of Claude Code.
These prices and tier terms are as of August 2025.
External Abilities
Claude Code has external abilities which are not available in Cursor.
Both Cursor and Claude Code can interface with MCPs, but Claude Code has a few
more external abilities.
Pull Github Issues and push Github PRs
Use command line to run commands while restricting some dangerous commands
like rm -rf
Per Project Settings
Optionally you can define a file called CLAUDE.md and this can contain your
instructions for a project which may have differing coding conventions, variable
naming, file structure, etc.
Key Takeaways
If you are working solo and value AI assistance, Claude Code provides the
best balance of power and usability. The learning curve is minimal if you are
already comfortable with terminal.
But as all things in AI... this may all change in the next week or days. I
suggest buying only a monthly subscription and canceling if you are not happy.
Looking Forward
The AI IDE space is evolving rapidly. By the time you read this, several of
these tools will likely have new features that change the equation. My advice:
Start with your use case: Are you prototyping, collaborating, or shipping
to production?
Consider your team: What tools do they already know and trust?
Think about lock-in: How important is vendor independence for your
project?
Test with real work: Do not just try the demos and build something you
actually need.