Amelia is a personal AI executive assistant that runs 24/7 in the cloud. She triages my inbox, defends my calendar, writes briefings while I sleep, and reports back over Slack. I'm building her in public.
A few weeks ago, my executive assistant was let go. The slot was open, the tasks were real — inbox triage, calendar defense, meeting prep, end-of-day digests, relationship maintenance. All the things that keep a CEO's week from falling apart.
Instead of hiring, I decided to build. Not a chatbot. Not a copilot. An actual assistant — one that operates on my real calendar, my real inbox, my real task list. One that runs while I sleep, in a different timezone, with my laptop closed.
Amelia is that assistant. She's software I own and control, running on infrastructure I pay for, with every decision logged and traceable. No black box, no vendor lock-in. Just Python, a Postgres database, and Claude.
Morning briefing at 6am — Gmail, Calendar, and Slack catch-up, prioritized and delivered to my Slack DM. Top-3 action items written to Asana.
Inbox triage at 7:30am and 5pm — reads my inbox, drafts replies in my voice, files newsletters and noise, flags decisions that need me.
Evening digest at 6pm — what slipped today, tomorrow's calendar overview, daily note prep for my Obsidian vault.
Calendar defense — auto-declines overflow, protects focus blocks, generates pre-meeting one-pagers 15 minutes before external calls.
Relationship maintenance — weekly nudges on contacts I haven't talked to in 60+ days. The kind of thing a great EA does that you never think to ask for.
Single-user, single-tenant, deliberately simple. No framework magic — just a scheduler, a database, and well-defined jobs.
Cloud loop proven. Heartbeat in Slack. Deploy pipeline working. Laptop closed, Amelia still running.
Morning briefings, inbox triage, evening digests. The daily rhythm of a great assistant.
Calendar defense, pre-meeting one-pagers, relationship maintenance, Obsidian knowledge graph.
June 2026 — Europe, 4 weeks. Six hours offset, laptop sometimes closed. Amelia runs fully independent. Pre-trip baseline set, escalation budget defined. Post-trip case study published here.
Because the question "can AI actually replace a knowledge worker?" deserves a real answer, not a demo. I'm running this on my actual life — my real inbox, my real calendar, my real relationships. If it works, the playbook is replicable. If it breaks, the failure modes are worth documenting.
This isn't a product. It's one person's bet that the tools are finally good enough.