A friend approached me recently to build a web app as a side project. Both of us have built and sold businesses before and had each felt the pain of having to hire people to write code and build out a company.
So the goal with the project was for him to come up with the big vision and design, and for me to build the app. The catch was that we had to leverage AI and SaaS based tools to build everything. No messing around with all the time consuming elements that come with building a new app.
I've been needing a project and ever expanding plethora of AI tools has been fascinating me so I agreed.
Introducing Cursor
The last few weeks I've been working with Cursor, an AI code editor. It's basically Visual Studio Code with an LLM built in. I can ask it to kick off a project using specific tools and then have it build out the features. Once I understood this, I was instantly in love!
We are looking to leverage Supabase (which I'll write more about once I get into it more) so I had it spin up a project and started having it create Edge Functions which are serverless functions.
It was so fun seeing it think through the problems I gave it and build out code. When you ask it to make changes, it steps you through what they are and you can choose to accept or reject the changes. It's pretty neat.
Is it any good?
Depends. Cursor is like a super smart programmer who is extremely forgetful about the codebase it's working on and doesn't remember what code it wrote vs what someone else wrote. A lot of time is wasted on it not remembering details. It's like it looks at the code fresh every couple of chat prompts and doesn't have a clear memory on what it did. This is where humans have a one up on it. Once they nail this, it'll be unstoppable. I am only using the default LLM integration and see you can hook it up to ChatGPT, Claude and so on so maybe it can be smarter.
For now, I think if you're developing a tool with a simple set of features or building something new, it's pretty neat. I can have it use specific libraries or frameworks and it can pretty accurately integrate with external APIs. Developing within a older, more complex tool could prove difficult but definitely worth a try and would depend on the codebase.
I've found Cursor gets me 95% of the way there with most things. It saves so much time whether starting something new or asking it to refactor how something is implemented, it's pretty good.
Writing unit tests are a huge pain and typically it goes around in circles (literally, it will retry the same approach multiple times...) trying to get the tests to work. It does eventually get there after a few "maybe take a step back and relook at it with fresh eyes" prompts.
Do I recommend it?
Yes! It's been great and allowed me to build out my new project with ease. I think it's worth a shot to see if it can improve your productivity. For $20 a month, you're definitely going to get a lot of ROI, not matter how much you're able to use it.
So the goal with the project was for him to come up with the big vision and design, and for me to build the app. The catch was that we had to leverage AI and SaaS based tools to build everything. No messing around with all the time consuming elements that come with building a new app.
I've been needing a project and ever expanding plethora of AI tools has been fascinating me so I agreed.
Introducing Cursor
The last few weeks I've been working with Cursor, an AI code editor. It's basically Visual Studio Code with an LLM built in. I can ask it to kick off a project using specific tools and then have it build out the features. Once I understood this, I was instantly in love!
We are looking to leverage Supabase (which I'll write more about once I get into it more) so I had it spin up a project and started having it create Edge Functions which are serverless functions.
It was so fun seeing it think through the problems I gave it and build out code. When you ask it to make changes, it steps you through what they are and you can choose to accept or reject the changes. It's pretty neat.
Is it any good?
Depends. Cursor is like a super smart programmer who is extremely forgetful about the codebase it's working on and doesn't remember what code it wrote vs what someone else wrote. A lot of time is wasted on it not remembering details. It's like it looks at the code fresh every couple of chat prompts and doesn't have a clear memory on what it did. This is where humans have a one up on it. Once they nail this, it'll be unstoppable. I am only using the default LLM integration and see you can hook it up to ChatGPT, Claude and so on so maybe it can be smarter.
For now, I think if you're developing a tool with a simple set of features or building something new, it's pretty neat. I can have it use specific libraries or frameworks and it can pretty accurately integrate with external APIs. Developing within a older, more complex tool could prove difficult but definitely worth a try and would depend on the codebase.
I've found Cursor gets me 95% of the way there with most things. It saves so much time whether starting something new or asking it to refactor how something is implemented, it's pretty good.
Writing unit tests are a huge pain and typically it goes around in circles (literally, it will retry the same approach multiple times...) trying to get the tests to work. It does eventually get there after a few "maybe take a step back and relook at it with fresh eyes" prompts.
Do I recommend it?
Yes! It's been great and allowed me to build out my new project with ease. I think it's worth a shot to see if it can improve your productivity. For $20 a month, you're definitely going to get a lot of ROI, not matter how much you're able to use it.
-Ben