Perspective · Writing from the team
Founder essays, engineering deep-dives, and short pieces on what we're learning from running this at our own company before we ship it to yours. New writing about every two weeks.
Everyone says they want to "find their shadow AI." The framing is wrong. You don't have a shadow AI problem. You have a productivity tool your engineers are already using on production code, and an obligation to govern it that the law just made expensive. Calling it "shadow" makes it sound like something to hunt. It's something to manage.
Read on LinkedIn →What runs locally in your browser and on your laptop, what reaches the dashboard, and what we deliberately do not do. Why a server-side classifier would have been the wrong design, and what the trade-offs of going on-device look like in practice.
Read the deep-dive →Northbeams runs on four surfaces. What each one sees, what it doesn't, and which to install first depending on whether your AI problem starts in a browser tab, on a developer's terminal, or inside an agent talking to your databases.
Read the breakdown →Most AI governance is browser-only. Your engineering org is somewhere else. This is the in-path proxy that sits between coding agents like Claude Code and the MCP servers they call, with per-tool policies and on-device argument redaction.
See the gateway →Coming next · MCP is the new SQL injection · What the EU AI Act actually means for your engineering org · 30 days governing AI across our own company · Why every browser DLP misses Cursor
About twice a month. Field notes, no marketing copy. Unsubscribe in one click.