Tokens or Toilet Paper?

Well, the day arrived when GitHub shifted Copilot toward usage-based pricing. Not surprisingly, the sticker shock hit hard and fast sticker shock hit hard and fast!

Looking back on the last few months, I am glad I started experimenting with AI-assisted coding when I did. Looking back at the charges for the past few months, my costs have skyrocketed, going from $1.50 - $2.00 to seeing $5.00 to $10.00 per day since June 1st.

As of right now, I am officially putting myself on rations.

A lot of what I have been building in the past few weeks is for personal endeavors and to explore AI Intelligent Apps. I have been doing a lot of coding by hand, but I have used Copilot in a couple of key areas, one of them being for Angular and React front-ends.

I unashamedly admit that I do not know front-end frameworks like Angular or React. I'm a back-end guy. But one way to gain exposure to those front-end frameworks - and maybe even use them one day - was to build a distributed (local) app with an Angular or React front-end.

The app I built with the Angular front-end was generated a few weeks ago. I have been moving along with that project, and I will likely grin and bear the cost because I really like the Angular front-end it has.

As for React, it just so happens that I created this very blog over the past weekend and decided I would kick the tires on a React front-end. Once the new GitHub pricing went into place, I quickly realized that building a React front-end was causing my usage to spiral out of control. I removed React and I switched to a Razor front-end because I know it. I can do Razor by hand, and any hang-ups I encounter now can go to Gemini.

I may explore some of these topics in depth at a later time, but I have to wonder what is going on behind the scenes. It's almost like Copilot has taken the side of the anti-AI coding army out there and increased prices to the point where it makes it nearly impossible for developers - and teams of developers - to be able to utilize AI Coding Assistance to its full maximum. Even if you do a lot more hand coding than AI coding, you still lose out on the ability to research and try new things from within the IDE. Part of AI coding is testing the agent's abilities, learning how to massage it, and gaining trust with it.

There seems to be little room for that anymore.