jOeCHEM - Carbon based chemistry made easy (for free)

In college, I used Chegg to help me succeed in my most challenging classes. Whether it was physics or my chemical engineering classes, Chegg helped me get my homework done (and correct) in a timely fashion and facilitated efficient studying where I could check if I got an answer correct, and if not, understand why I got it wrong. I had this incredible group of engineering friends in college, super smart and also super hard working, and we frequently joked about how we would be so lost–and frankly screwed–without Chegg as our guiding light. It was our super power, the best money we spent each year, as long as we wielded it responsibly aka we avoided blind copying of solutions and made sure to truly internalize concepts and the problem solving techniques at work.

In short, Chegg enabled me to work both smarter and harder, accelerating my learning and ability to prep for exams. Instead of fighting through the initial friction to get to the part where I’m actually learning, I would quickly get to solving physics problems, checking if I arrived at the right answer, and if not, diving into where I went wrong. And I loved this studying feedback loop because it rapidly got me ready to solve questions on exam day, which was the goal in the first place.

So when I read Thomas Ptacek’s article about AI assisted coding, this part hit hard:

An LLM can be instructed to just figure all that shit out. Often, it will drop you precisely at that golden moment where shit almost works, and development means tweaking code and immediately seeing things work better. That dopamine hit is why I code.

The current state of AI assisted coding tools feels like that same Chegg super power I experienced in college. One thing that has always been on my “programming TODO lists” was to make a TUI with Go and Bubble Tea. I enjoy writing Go, so there was no particular reason for the language/framework choices other than wanting to flex my Go muscles and build a type of application I never have before. I’m an infrastructure and backend type of person, so working with any type of interface/UI horrifies the right half of my brain. But armed with my Claude Code subscription, after 12-15 hours across three days of working, I had a working Spotify TUI!

The amount of time it took to go from 0 lines of code to something that handled the OAuth 2.0 flow with Spotify (with automatic token refresh!), had reasonably nice looking Bubble Tea panels on the terminal screen, and played songs was honestly just a few hours. I know people make claims like this all the time, but I swear–I had a hideous (it looked bad, trust me) but working TUI in a matter of hours. The bulk of the time I spent on this project was tinkering, experimenting, and trying new things out and either succeeding or failing and having Claude figure it out, and then me exploring more. It was just like using Chegg back in college: I didn’t get bogged down by the terrors and pitfalls of OAuth 2.0, nor was I burdened with coming up with an initial interface concept. Either of those two things alone may have intimidated me from this undertaking.

Rather, in a short amount of time, I was spending my time on my intended goal, exploring the Bubble Tea framework and messing with different toggles to build my understanding of how it worked. And at some point, Claude can’t figure everything out, and I had to put on my debugging/researching hat to identify how to get the song progress bar working and persist on the overall TUI model’s state using Bubble Tea’s Every function. And resolving that progress bar issue was an affirming sign that I had not just cluelessly prompted my way to a prototype. Just like with my studying, I had made sure to invest the time to understand what had happened, ultimately accelerating my understanding, and having fun while doing it!

Now, a skeptic could point out that I had code generated over parts of this project I myself should’ve implemented. While there’s some merit to that, I would also say that in three days (of after work time, mind you) I would’ve barely gotten the Spotify OAuth 2.0 flow to work because OAuth 2.0 sucks. I don’t live and breathe OAuth 2.0, and the point of this personal project was to make a TUI–it wasn’t to struggle through getting OAuth working. Instead, I got an efficient refresh on OAuth 2.0, a functioning TUI prototype, and even a GitHub Actions setup that on a new git tag push builds and publishes binaries for Linux arm64 and amd64, MacOS arm64 and amd64, and Windows!

Is my Spotify TUI going to catch fire and get downloaded thousands of times? Will it start racking up GitHub stars? To both questions: Hell no. It’s not feature-complete in a lot of ways, not like https://github.com/Rigellute/spotify-tui is. Sure, my TUI doesn’t automatically play the next song after one finishes (oops, I should fix that), but it works. I actually took the time to make it! In a pre-AI assisted coding world, I would’ve kept kicking this TUI-can down the road, making continual hollow promises to myself that I would make the time for it some day. But instead, in this reality, I have my first TUI under my belt, and next time I set out to make one, I’ll be even better at it.

Chegg helped me make sense of and excel at electricity and magnetism during my time in Physics 2, topics that felt impossibly confusing. For the first time in my career, areas such as frontend development don’t feel out of reach for an infrastructure/backend person like me, now that I can efficiently explore and learn. But sorry, I gotta go: Time to make sure my TUI can automatically play the next song.