What I Did
- Squash-merged monorepo PR#1 (37 commits, 151 files, +7131/-2337)
- Staging data migration + verification complete
- Fixed CI pipeline issues (type-check/test temporarily skipped, tracked in an issue)
- Defined Phase 1-3 production release plan
- Compared subdomain vs path-based routing using 3 different AIs
- Ordered a multi-tenant architecture book
What I Learned
Subdomain vs Path-Based: What 3 AIs Agree On
I fed the same prompt to 3 different AIs to compare their takes.
What all three agreed on:
- Start with path-based routing in the early stage
- Path → subdomain migration is easy; the reverse is painful
- Hybrid (LP/blog = path, app/admin = subdomain) is the realistic answer
Where they differed:
- AI-A: Leaned into security/compliance (FISC, PCI DSS)
- AI-B: Emphasized business phases (0→1→10→100)
- AI-C: Framed it as "cognitive load" vs "SEO asset maximization"
Your AI reflects your own biases. Cross-checking with multiple AIs is a healthy habit.
Value vs Action: History Table Design Patterns
Learned from a mentor about storing deltas instead of snapshots.
- Value approach: "Balance: 1800 yen" → no idea why
- Action approach: "+1000, -200, +1000" → full trail of what happened
- In practice, you use both: main table (Value) for fast reads, history table (Action) for full audit trail
Bank ledgers, Git, Redux, double-entry bookkeeping — all the same idea.
Little Snitch for Linux
The beloved macOS network monitor now runs on Linux, using eBPF for kernel-level outbound traffic monitoring. The team honestly states it's "built for privacy, not security" — a distinction that matters.
Reflection
Don't stop at asking one AI. Today I practiced cross-validating across multiple sources. The Value vs Action pattern from a whiteboard session needs to become code, not just knowledge. That's the next task.