Hello again.

I have official graduated from the Class of 2026, and will soon head to Dartmouth College. Let's take a look back through memory lane and see what I've learned.

It's Hammertime

I've been watching a lot of F1 and MotoGP lately, mostly the former. I was interested in Marc Marquez' redemption story with Ducati last year, but now I'm interested in watching Ferrari and Lewis Hamilton get their next world championship. I think that as I learn more about the history of F1, their drivers, their stories, and the kinds of inspirational stories that they put out, it helps me learn about not giving up, continuing to push forward, and the deep intricate details about how everything works. F1 is both a team sport and an engineering sport, all led by the driver themselves with their team principal in the back.

It's also really fun to watch while I type away at my computer.

Claude vs Codex

My relationship with AI Coding has definitely evolved over the last few weeks.

I started using Copilot more this year as I began working on The Earth App. It definitely helped as I migrated to a PHP backend, worked through UI bugs, and implemented new backend features in places where my knowledge about PHP and TypeScript waned. I had always been using code completion tools and suggestions, but never the full agentic chat mode until recently. I think it's really interesting on how to use; on one hand, you have a bunch of vibe-coded slop that likely won't work. On the other hand, that is slowly becoming not true anymore with the evolution of Codex-5.3, Sonnet 4.7, and Opus 4.8 (as of writing this); the models in the next few years will likely be way more powerful than what we have right now, or they may plateau with consuming near all the documented information that we could possibly synthesize.

It is also extremely unprofitable and I believe the bubble is going to burst. The second GitHub nerfed the student copilot plan into the ground, my initial suspicions about AI were confirmed. Not enough people are using AI to justify the kind of scaling vs pricing model that they currently have. As of writing this post, GitHub still is disallowing new Copilot subscriptions. And, they are continuing to post blog posts about more AI models continuing to be removed from the list. After Sonnet 4.7, they removed Codex-5.3 a few weeks later from initial selection, and you basically had to learn how to engineer Auto for your own complex use case. I eventually just decided to switch to Claude Code itself, which came in nicely when I finally received my student email from Dartmouth, which comes with Claude Enterprise access.

I think the burst is going to be much less dramatic than what people make it out to be. We will always need humans in the loop, no matter what kind of activity or purpose we're doing. Otherwise, you end up with a budget or crappy design that nobody likes. The bar will just go up: professional movies, designs, and websites will increase in UI/UX value, artistic intent, or just overall coolness in general. Slop will still be a thing, we'll just increase the bar for what slop looks like. We'll place more value on human-done things.

Think about it like this: we place a lot of emphasis on distance traveled, it just depends on how you got there. Running a marathon is pretty cool. Biking it is good. Driving it is... fine. Doing it by plane would take a short Instagram reel.

Moving On

Got some summer projects lined up: as of writing I'm finishing up the iOS mobile build for The Earth App. The Android one would likely take little to no effort, considering that I don't have a Play Store Developer Account just yet, and want to gather some feedback with my iOS build with different people before release on June 14th.

Have two other projects lined up on The Earth App for summer builds, as well as finishing smoke , a self-hostable customer support app I'm working on with Nuxt and Cloudflare Email routing with Cloudflare Workers. Fun stuff.

See you all soon!