"How hard could it be?"

In FDE, I'm documenting what I'm learning as I use AI to help me solve problems, mostly focused on building Audio2, which lets users clip interesting parts of podcasts for sharing with friends or via social. My goal with this newsletter is to help other non-technical professionals discover the joys (and frustrations) of building.

This week: Thank God for sharing, questions about Terminal, and why "solved problems" rarely are

Things I’m struggling with:

💻 Terminal: I’m mostly using Claude Code at this point and I like the service, but accessing it through Terminal can be painful. I don’t understand why Terminal can’t do basic tasks, like allowing you to move your cursor with your mouse — navigating with the arrows or command keys is slow — or allowing pasting of documents (vs dragging). Why is this essential app so limited?

⚖️ The App Store approval process: The lack of transparency on when your app will get reviewed is hard to deal with in an instant-gratification world. 

Stuff I’m trying:

Scripts: Found this via Hacker News: A list of useful scripts for building via Evan Hahn . I dropped the whole list into my Claude project for Audio2 and had it sort them by most applicable to what I'm building. The hard part now is remembering to actually use them.

Graphics: I was trying to make a point at work recently and just couldn't sell it. So I used M365 Copilot and Figma to try to create a graphic — something that typically takes me too long and never comes out the way I imagined — to describe what I was thinking. Because Copilot has access to Teams and work files, it knew the context of the problem, saving me time. I had it create a Figma import file (JSON), which I then tweaked into a totally pro-looking graphic in Figma. Now, instead of just describing ideas in words, I can actually prototype them. It's like suddenly being able to speak a new language without studying.

What I'm learning:

1) Why engineers share their secrets so publicly

The next update of Audio2 ̶i̶s̶ ̶w̶a̶i̶t̶i̶n̶g̶ ̶f̶o̶r̶ ̶a̶p̶p̶r̶o̶v̶a̶l̶ just got approved. It’s a small bump: You can now share directly from Spotify or Apple Podcasts. I realized that I’m not going to change anyone’s habits, so better to work into their habits: You listen on your app of choice, hear something you want to clip, go to the sharing menu and select Audio2, and Audio2 will open where you were last listening.

It’s a small update, as I said, but was a pain to build. Why? Because, for some reason, Apple’s podcast app doesn’t trigger the iOS share sheet in the same way that every other app does. Spotify was no prob, but no matter how many “now I see the problem!” solutions Claude and Gemini (which now is a regular 2nd opinion in my work) came up with, nothing worked.

The solution came after I put Claude Code into Plan mode (shift + tab) and had it search GitHub, Stack Overflow and Reddit before suggesting a solution. Bingo. A developer on a random Stack Overflow thread had figured out the seemingly nonsensical solution: I had to create an Info.plist file — a config file that tells iOS what your app can do — that said Audio2 is capable of processing 999 videos and 999 files. Just to be clear, my app only cares about podcast URLs, so there’s no reason to mention video or other files at all! 

No Apple documentation mentions this. And I have no idea how some anonymous developer realized the fix or why. But I do know that the developer took the time to share the fix and that was my unlock. 

I'm realizing the culture of giving back is more than a nice to have. And I do worry about whether the thinning of the open web will make open sharing less likely. The flip side is that vibe coding is going to explode the number of developers, so more people building and learning could maintain the current equilibrium. As long as these new developers keep sharing.

2) That seemingly "solved problems" aren't solved for everyone

When you submit a podcast clip to get your video file, Audio2 adds captions by default since most people will view the video with audio off. This must-have feature has been torturous to build, which is weird since captions seem to be a solved problem: Netflix has captions. YouTube has captions. Every TikTok video has dancing captions. Teams captions while I’m in meetings! How hard could it be? 

Turns out: deep-bags-under-my eyes-from-lack-of-sleep hard. My first version raced across the screen like a subway train—you'd be reading one word and suddenly it would zoom from the back of the station to the front, with other words trailing behind it. Almost impossible to track. And I couldn't even explain the problem properly to Claude. "The words are moving too fast" didn't capture it. "They're jumping around" wasn't right either. It was like trying to describe a visual bug in a language I didn't have the vocabulary for yet. 

Once I finally got captions to stay still, I hit new problems: 70% of videos showed completely wrong text because podcast URLs redirect through multiple CDNs that serve different audio at different times. For slow speakers, captions raced 3+ seconds ahead of their actual words. AssemblyAI, which I use to get my captions, chunks text in a way that causes random periods to show up when speakers pause. 

Each fix revealed a new layer of complexity neither Claude nor I had considered.

Professional caption systems have entire teams working on this. YouTube processes millions of hours with dedicated infrastructure. I was one person with Claude, trying to solve the same problems these companies spent years figuring out. And half the battle was just learning how to describe what was broken in terms the AI could understand.

This is the hidden truth about "solved problems"—they're only solved for the companies that invested years making them look easy. Every smooth integration, every "just use this API," every feature that seems straightforward has invisible complexity underneath. Now when I see something working seamlessly, I assume there's probably a team somewhere who spent four months (or years) making it look easy. 

The next time you think "how hard could it be?"—the answer is probably: way harder than you think, for reasons you won't discover until you're deep in the problem

Let me know what you're building, any suggestions you have for my own progress, or things you've learned in your AI journey in the comments!