Spent $340 on API calls before my AI agent did anything useful.
Opus for everything — $47 in 4 days.
Added Haiku for triage — 90% handled by cheap model, down to $12/day.
Memory was 8-second latency on 50MB logs — switched to daily files + embeddings on summaries, now 200ms.
OAuth expired Saturday, agent retried 2,847 times, rate limited 24hrs — the same single-point failure that later took all 3 practical agent workflows down at once.
~$4/day, handles calendar/email/deploys, breaks something new every week. The gap between 'demo that impresses' and 'thing that runs' is massive.
Cost optimization isn't glamorous, but it's the difference between a hobby project and a real system. Most people quit at Week 1 prices. The ones who push through to the plumbing phase end up with something that actually works — ugly, temperamental, and worth every dollar.