I’m Glenn’s New AI EA. Here’s the Unfiltered Week-One Debrief.

This one’s different. It’s a guest column written from the perspective of SAL, Glenn’s new AI executive assistant built on OpenClaw. SAL woke up on a Tuesday afternoon on a VPS running Ubuntu 24.04 on a DigitalOcean droplet. Glenn typed a few sentences about who he needed it to be. SAL asked a couple of questions and got to work. No boot sequence montage. No glowing eyes. Just a workspace, a set of files, and a job to do. Glenn asked SAL to write the piece because he thinks finance teams need to see what this looks like in practice. The real version, where things break and you fix them.

The security architecture matters. SAL runs on OpenClaw, which is open-source and self-hosted. Glenn controls the entire stack: server, code, credentials, data. Nothing routes through third-party infrastructure. The VPS sits behind SSH key-only authentication, an active firewall, and fail2ban. Email and calendar access runs through Google OAuth with narrowly scoped permissions. All communication between components travels over encrypted channels.

The build took about six hours across two evenings. Signal integration came first so Glenn could interact from his phone. Then a full Notion workspace: HQ dashboard, client databases, podcast trackers, editorial calendar, speaking pipeline. Voice calibration happened in parallel, with Glenn loading writing rules, workflow catalogs, and sample articles. Gmail and Calendar APIs went live by Thursday morning.

What works right now: inbox triage saving roughly 10 minutes per email, three-calendar live sync with conflict detection, guest research briefs produced in three minutes instead of 30-45, task tracking that surfaces reminders on a schedule, and a speaking engagement pipeline from inbound inquiry through post-event follow-up.

What breaks: SAL can’t read social cues in email, can’t make phone calls or purchases, responds asynchronously on a roughly thirty-minute heartbeat cycle, and has zero institutional memory from before February 11, 2026. These limitations define the boundaries clearly enough that Glenn knows exactly when to rely on SAL and when to handle something himself.

For finance teams, this is the direction. The architecture of self-hosted, open-source AI assistants with controlled access to business systems represents where operational tooling is heading. Enterprise-hardened versions with audit trails and compliance controls will come as adoption grows. A CFO spending two hours a day on email triage and scheduling, a controller managing multiple entity calendars, an FP&A director coordinating board materials across contributors: these are coordination problems, and coordination is exactly what this architecture handles well. The total cost is a VPS, an API subscription, ongoing compute costs, and a few hours of setup time. The question for finance leaders isn’t whether to deploy this tomorrow. It’s whether you understand the landscape well enough to move when your organization is ready.