Last week’s dev diary focused on a month of rapid experimentation and cleanup. This week was quieter, but no less revealing. Behind-the-scenes infrastructure, a small project that turned into a useful tool for someone else, and an entire evening wrestling with a problem that still has no clear solution. The common thread? AI has lowered the barrier to creation, but the last mile—deployment—still demands technical know-how most people don’t want to learn.
The final mile where most projects stall
A marketing teammate showed me a beautifully designed event page they’d built using AI. The colors matched the brand, spacing was correct, and it fully complied with our design system. Then came the real question: “How do I get this live?”
What surprised me wasn’t the design—it was the blocker. The hard part, building the interface, had already been solved by someone with no technical background. What tripped them up was publishing. A trivial Git or server task for any developer, but an impenetrable wall for someone who’s never touched those tools.
I created a GitHub Pages repository where each folder automatically becomes a subdomain under our main site: ourdomain.com/event-name. Fifteen minutes later, the page was live.
This moment stuck with me. AI has democratized interface creation, but infrastructure still assumes knowledge most users don’t have—and likely don’t want. The role of developers isn’t shrinking; it’s evolving. Increasingly, we’re not just building, but finishing the journey for others who’ve already done the creative heavy lifting.
The week that never makes the demo reel
This week was filled with infrastructure work, permission wrangling, and subtle state inconsistencies—tasks that rarely make it into slide decks or product showcases. One bug haunted us for days: a silent failure during login. No exceptions, no error messages, no obvious logs—just a user stuck in a loop with no explanation. When we finally traced it, the fix was three lines of code. The hard part wasn’t the code; it was reaching that insight.
We also ran a heavy QA cycle on a complex screen where state flows in from multiple sources and edge cases only emerge when everything is combined. We had automated tests in place, but they weren’t enough. Automated tests catch what you anticipate. External QA finds what you never imagined. Two forms of coverage, neither a substitute for the other.
The rest was cloud infrastructure—permissions, unexpected costs, and configuration surprises buried in settings no one expected to touch. Infrastructure work has its own rhythm: you think it’s a two-hour task, it becomes a half-day saga, and by the end, you’re testing again just to be sure.
Giving game characters Portuguese voices—one dead end at a time
I’m building RPGTeller, an engine for gamebooks—interactive stories where choices shape the narrative. This week, I explored adding voice narration with distinct character voices.
My first stop was Kokoro, an offline text-to-speech engine. It’s appealing for local-first projects: no API calls, no per-character charges, and it runs entirely on your machine. The catch? Brazilian Portuguese came out robotic and unnatural. Functional, but far from immersive.
I switched to ElevenLabs. The quality improved dramatically. I could assign specific voices to each character—the narrator, Svetlana, the warrior—and imagine how it would sound in-game.
But ElevenLabs is a paid API, charging per character. Before committing to an ongoing cost, I want to know if acceptable quality is possible with a local solution. For now, Kokoro was too rough, and ElevenLabs delivered but relies on the cloud.
I haven’t solved it yet. More local options still need testing, especially those that balance quality and offline capability for Brazilian Portuguese. The search continues—but the journey itself is part of the process.
The first diary covered a month of ups and downs. This one covered a typical week. If you’d like to follow along, you can find the next entries on the platform.
AI summary
Yapay zeka sayesinde herkesin kolayca oluşturabildiği arayüzler, yayınlama aşamasında nasıl beklenmedik engellerle karşılaşıyor? Altyapı bilgisinin ve QA’nın önemi.
