I build Pandaz Apps with a lot of help from AI these days — though I spent many years before that writing every line by hand. Panda Calendar itself was actually born during COVID-era remote work: when the daily commute disappeared, those recovered hours finally gave me the time to build it. These days I have chances to watch several coding AIs in action, and right now — by a pretty wide margin — Claude Code is the one I recommend. Let me write down why, partly as a record for future me.
The Tools I've Actually Touched
My main driver is Anthropic's Claude. My son is in university and has a Gemini AI Pro license, so I've watched him work with Gemini over his shoulder. At my day job we have a ChatGPT Enterprise license, which has given me a chance to try Antigravity and Codex as well.
They're all capable in their own ways. But for the kind of hands-on coding I do on Pandaz Apps, Claude Code has been the easiest to get results from. I use it as a VS Code extension, which puts it right next to the files I'm editing — no context-switching, no copy-pasting into a separate chat window.
Remote Control Saves Me on Busy Days
One feature I didn't expect to rely on as much as I do: the remote-control piece. When I have to leave the house in the middle of a task, I can check in from my phone, see where things stand, and give follow-up instructions without sitting back down at my desk.
That sounds small, but when a build is running or I've asked for a refactor that will take a while, being able to course-correct from a café or a train platform is genuinely useful.
Sonnet Up to 4.5, Then Opus
Through the 4.5 generation I used Sonnet. This wasn't about saving tokens — it was about the balance between response quality and the time I was willing to spend. For what I do, Sonnet at 4.5 was the sweet spot.
I had been quietly tempted by Opus for a while, but the full switch happened at 4.6. Part of the push was that Sonnet had gotten a touch slower — but honestly, I think I would have switched anyway.
What Opus 4.6 Changed for Me
The thing I noticed first with Opus 4.6 was how much less I had to walk things back. It stopped doing the two moves that used to eat my afternoons: starting a task completely over when a small nudge would have been enough, and breaking something unrelated because it hadn't fully considered the ripple effects of a change.
Fewer do-overs means my actual progress per hour went up, even if the per-request speed was similar.
And Then Opus 4.7
Opus 4.7 took another step forward, and the thing I'd single out is its grasp of the application as a whole. It doesn't just answer the question in front of it — it factors in how the app is put together and will come back with improvement suggestions that actually fit the codebase.
The trade-off is that the actual edits feel a little more deliberate — maybe a touch slower. But the investigation phase has gotten faster, so net-net I don't notice the slowdown. If anything, I trust the output more, because it reads the room before it writes.
The Best Part: Family Can Join In
The thing I've come to love most about developing with AI isn't the productivity numbers. It's that the door is open to my family now, too.
My son watched me work, picked it up, and built FamiSke himself. A few years ago that would have been a long road — setting up a toolchain, learning a framework, pushing through the boring parts. Now the boring parts are shorter, and someone with an idea and a bit of patience can turn it into an app.
That's the part I didn't predict. When I talk about why I chose Claude Code, I start with "Opus 4.7 is great." But what really matters is that tools at this level make building something reachable for more people — including the ones I share a dinner table with.
See what we've built as a family.
Explore Pandaz Apps