Launch — My Wife Opens Pepper
There was no grand scene.
No App Store release, no press, no first-user graph. I pushed a build to TestFlight and sent Soyeon an invite link. Install. Log in. Done.
Right now, two people use Pepper. Me and my wife.
So this is what launching is
Months of talking about STATE A/B/C, dragging triage from 33% to 90%, catching three silent bugs — and "launch" was the quiet scene of my wife opening her phone in the morning to check the briefing.
Watching Soyeon use it those first few days was strange. The inputs I'd imagined from T-001 to T-500 in the benchmark, and the things my wife actually says, were different. At the end of EP 11 I wrote "the real number will reveal itself in use" — and that "in use" is now. Real usage logs have started piling up, tagged with triage_confidence. Anything below 0.8 confidence surfaces for review. Pepper now learns from what my family actually says, not from cases I imagined.
Honestly, what I didn't do
In EP 10 I declared: "STATE B — the moment Pepper writes its own code — is the real finish line for Phase 0."
Even now, at launch, Pepper still can't write its own code.
Because I moved the finish line in EP 12. I decided to put STATE A (looking after us, doing what we ask, remembering us) into the family's hands first, and pushed STATE B (self-evolution) and C (human intervention) to Phase 2. I still think that decision was right. Rather than a self-evolution that never ships, a briefing that arrives at 9 a.m. today was what my family needed more.
So this launch is half a launch. Pepper looks out for us, does what we ask, remembers our family. But it can't yet grow on its own. That sentence is still in the future tense.
Three months ago
In EP 0, I wrote that I didn't even know what GitHub was. That I first learned what a commit was, what a push was, building kimeunsoo.xyz.
Three months have passed. Now on my phone — and on my wife's phone — sits one app running a monorepo, a router, six-way triage, three layers of memory, a pg_cron scheduler, and Vault embeddings. I built it. More precisely: I built it by leveraging AI.
I reread the declaration from EP 0.
"It's a raw, unfiltered record of a non-expert leveraging AI to build something. It might be rough. That's fine."
There's plenty that's rough. lookup is still at 73%, the silently dying tokens aren't fully exposed, the planner still says "coming soon." But the app runs every day on my wife's iPhone. That much keeps EP 0's promise.