Скачать или смотреть Passing Out with AI - Cursor Plan Mode Success - Token Use Tracking

In this video, I passed out using the new Plan Mode in Cursor while implementing token use tracking. I wanted to track how many input and output tokens were used, and then using the current token price to calculate my usage.

I've had Plan Mode for a few days now, didn't know it was a "new" feature until I saw a news release in my email. This was a good time to use it, since I knew it would be a relatively easy feature to implement, and can be implemented across all my calls that I built previously. By now, the application is in a relatively "mature" state, with some clear standards in place.

What I didn't expect is the time it took. It was the end of the day, and my eyes were getting heavy. So, during those long gaps waiting for the LLM, my eyes decided to close.

I'm almost at the finish line with the extraction portion of my application. I suddenly had a thought, why not build in some token tracking feature for each run? Since some parts of my application does multiple calls to the LLM and using different models, it would be great to know the cost associated with each call. That way, I could have some better analytics with tracking costs associated with a worksheet that gets uploaded.

I'm not an advocate of one-shotting your problems, but i'd say this was a successful one-shot.

I have a variety of thoughts when it comes to problem solving with LLM's, but they revolve around a single concept of clarity vs ambiguity. This leads to different experiences, but at each end of the spectrum, you get the "sinusoidal wave experience" and the "exponential curve experience". I'll definitely make some videos on this, as this is a big part of my philosophy and my understanding of building with LLM's at the moment.

-------------------------

Hello! I'm building an MVP product with a partner in Hong Kong that helps teachers and tutors create more engaging and fun content for kids in K1 to P3.

My partner has zero coding experience, but with tools like Lovable, she's able to move along with progress. We meet every Monday and Thurs, so i'll attempt to get those streams online! Its an awesome time to build! In the past, she will have to work with a designer or talk to somebody to help her get her thoughts into something visual like a wireframe, that used to be a 2-3 day process.

I've been developing for 20+ years and coding with AI tools for the past year and a half. LLM's are like any tool, they have their pros and cons. I want to talk about my philosophy, what i've learnt, and how I think these tools can be used well. I'm just learning like everyone else! But would like to definitely share my journey so far.