Cursor coded 210,000 lines of code for me in May. The cost? $40

Sitemap## Realworld AI Use Cases

I wanted to have a publication that helped share tips and tricks from my consultancy where we implement ai for companies like the Olympic team, the Stock Exchange and other B2C companies.

Isn’t AI incredible?

210,000 lines of code and I have been under the weather in May and so that is lower that normal.

I spent $40 in May, $20 for the base plan and then $20 for some additional requests.

Only $40. I can’t fathom how valuable this is.

The interesting thing is that I dropped my total spend from close to $100 to $40 and ==I actually increased the total number of lines from last month==.

This aligns to my intuition, I “feel” like I am getting significantly more efficient at coding using Cursor and it is also rare for me to feel frustrated in a day.

Not a Medium member? Keep reading for free by clicking the link here.

So here are the tips I want you to remember

The first one:

If you can’t imagine it, STOP, neither can the AI.

If you can’t articulate what’s on your page, then don’t prompt Cursor to build anything.

If you can’t imagine how it will look, then don’t prompt Cursor to build anything.

If you don’t know how the data will flow, then don’t prompt Cursor to build anything.

Yes, I am a fan of vibe coding, but I have come around to the conclusion that we need to now graduate beyond pure vibes.

So do this instead

Go OUT of Cursor. To ChatGPT/Claude/Gemini and basically brain dump and get it to ask you a whole bunch of questions, clarifying your feature. Also specifically tell these LLMS not to write any code or worry about anything technical, only to write in english (they are over eager on their artifacts.)
Then go back to Cursor …. but still don’t build anything!
Paste your text from step 1 and then ask Cursor to clarify more questions.

See below for an example of 4 questions Cursor answered out of 13 in total.

I want you to spend far more time answering questions going forward than you think.

We still aren’t done.

When it is implementing, still critique, still add things.

I often clarify about coupling things up and making sure it doesn’t try to do too much

Watch this talk with one of the members from the Google Labs team, he talks about how their prompts for coding are pages and pages long.

The Result?

You will drastically reduce the number of bugs and rework that you have to do. I often end up with a whole bunch of one shotted features now.

If this still isn’t working then please write me a message and let me know, as I guarantee you that you will be doing something wrong and I would love to help.

Usually my requests result in a message like the one below and everything just works.

It’s nice seeing this all

Prototype the UI with v0

I do a lot of UI prototyping for features that I am not clear on with v0. Let’s take an example of a system that is an infinite canvas that you could create cards and text.

I will spend 5 to 10 minutes playing with my idea on v0 for these types of ideas.

I also like to play around with thematic UI designs. Like giving it a minimalist look and seeing what that looks like.

v0 is great for doing thematic UI designs

Personally I find v0 better at smaller lightweight tasks and pages so I like to keep it really small.

I keep my v0 files really small

That also means that I can paste the page into Cursor as context and then it really understands what I am doing and the migration process isn’t really complicated.

This all helps you to visualise what you are building.

Often times I put it into v0 and my idea that I thought was cool, actually looks lame as and so I just saved a whole bunch of time!

The 5 minute test

For the most part if you are going more than 5 minutes on a feature and going backwards and forwards then you most likely have made it too big.

Go back to your last commit and start again.

You shouldn’t be wrestling and wrangling the AI. You shouldn’t be having long time out sequences and large grepping sequences. I’ve come to see all of these things now as you not optimising the AI enough. If you find your self frustrated, swearing in CAPS, and things constantly breaking then you are doing something wrong most likely and it’s better to just assume that you are.

Of course there is going to be exceptions to the above, but a flow for adding new features is, explore your options, write planning documents, answer the AI questions and then have it implement it. From there, you just edit the minor bugs.

The refactoring method that works

I got an awesome comment from Rafa recently about how he used rules on the recent refactoring post I did.

That’s great to hear that these tips are resonating with the audience and I really liked the last line of how Rafa’s repos have never been so well maintained.

The craziest thing to me is that the AI is only going to get better from here

$40 for 210,000 lines of code in a month is just ridiculous.

It is already at the stage where the price to value is astronomical. We have all been given this absolute miracle of a tool to use and it’s just bizzare that it is so cheap.

Remember I work for large enterprises and global brands and this code is used in billion dollar businesses and it cost me $40! The tech teams of some of our clients aren’t even using AI yet! It is just madness!

I’ve done a breakdown of pricing of Cursor before but I just have to keep posting about how good this is because it has gotten even better this month I think with their recent pricing and request update.

Also Claude-4 seems to have really solved a lot of the over eager problems that Claude 3.7 had. For an update on that model front, with this planning framework, I now primarily find that I am using Claude 4 and unless it goes wildly off base, I have not needed to rely on the strict implementation of 3.5 yet.

What is your Cursor Usage?

I am interested in your usage and if you go to your Cursor account you can see it. Have we got any power users here? If you have done more code than me, can you also share a tip that has worked for you.

Other Tips & Tricks

Last published 18 hours ago

I wanted to have a publication that helped share tips and tricks from my consultancy where we implement ai for companies like the Olympic team, the Stock Exchange and other B2C companies.

I’ll help you code with AI. Cursor tips & Business Strategy. I run a company doing AI for the All Blacks, Olympic Team & the Stock Exchange www.cubdigital.co.nz

Responses (14)

Talbot Stevens

What are your thoughts?

I have some trouble with the idea that 210.000 lines of code is a good thing. Who is going to maintain this mountain of logic? What does it even do underwater? How secure is it? What if you have to scale it up?

Some of my best days of coding are when i significantly reduce the number of lines of code.

Over my many years of programming, I’ve learned that bragging rights should be reserved for how few lines of code are required to achieve a result, not how many. Most enterprise projects I have seen fail this test spectacularly, using far more code…

Yes, I align with this as well. I find myself spending a lot of time with both ChatGPT, Claude, and Perplexity asking questions and guiding the conversation around logic paths, inputs, outputs, industry standards, edge cases, etc to make sure I have…

Recommended from Medium

[

See more recommendations

](https://medium.com/?source=post_page---read_next_recirc—ecacb5696f5f---------------------------------------)