Hudson McGough

Speed was never the benchmark. Ever since building with AI became mainstream, speed has suddenly become a focus. The faster, the better. But why? For startups where timing and outpacing your competitors is everything, sure. But for bigger companies, shipping crazy fast was never actually the bottleneck. Now that AI can build things quickly, it's somehow become the centerpiece of building software. It’s every AI companies' main selling point, and now the thing everyone optimizes for. The issue is nobody's asking whether it should be.

One of the most rewarding things I've been a part of this year wasn’t for a startup. It was for a sushi restaurant. While in Cape Town, my friend Luke Braxton and I built something for Codfather, a sushi and seafood restaurant right next to where we were staying. When we pitched the owner on an RFID system to track plates off his conveyor belt, he told us he’d wanted it for years but could never get around to it. We built a low-cost prototype with all the software in place. Once it was working, we handed everything off along with a hardware implementation guide so they could take it from there. If you’re ever in Cape Town, visit Codfather and look for the new system!

I went to Cape Town almost two months ago. I just landed back home. I went there for an internship and study abroad experience, but I left with a lot more than that. I learned a lot. With a desire to explore, I attacked South Africa wanting to seek discomfort and get some hands-on experience. I met many incredible people, worked with startup founders, built real products, and even helped local businesses optimize their craft (more on that soon). I especially want to thank Nick Hoare and Richard Braithwaite for being such amazing bosses and co-founders. I’m happy to say that although I’m leaving Cape Town, my work at SVYP isn’t over yet. I went from being an eager kid hungry to learn about the dynamics of a startup to an engineer who built real products and felt like a co-founder by the end of it.

Be meticulous. As a data & ai engineer at SVYP, a big part of my job this week has been making complex data feel simple. So clear that clients can immediately see what they're missing, why it matters, and how SVYP can close that gap. A lot of that work has come down to prompting. Specifically though, prompting claude until I get exactly what I want, not just something I can live with. There's this idea that the fewer prompts it takes to get a "good enough" result, the better you are at using AI. But the benchmark shouldn't always be efficiency. It's how close your output is to the vision inside your head. Sometimes that takes manually editing the code, or going back and prompting little changes again to get things just right. Every part of your build should be intentional. That's why the extra prompt is worth it.

Today I built an MCP server for my favorite to-do app. A few weeks ago, I started using Habitica (by HabitRPG) to track my to-dos. If you're not familiar, it basically gamifies your life. I've been building a second brain dashboard that consolidates my to-dos, mail, calendar, and more. Since I migrated my task tracking from notion to habitica, I looked for an existing MCP connector. They were pretty much all paywalled, so I built my own. If you use habitica, try it out. If not, I'm putting you on, trust me 🙏🏽 Github link in the comments!