Most conversations about AI assistants focus on capability: what can it do, how many tools can it use, how far can it act without you. Alden starts from a different question, not "how much can it do," but "how does it earn the right to do it." That's what we mean by governance-first AI.
An assistant that can fill out forms, manage files, and click through your browser is only genuinely useful if you can trust it with that access. Capability and trust are usually treated as a trade-off: more autonomy for more usefulness, less oversight for more speed. Alden is built on the premise that this trade-off is a false one. An assistant earns the right to act by asking first, not by acting first and explaining later.
Every action Alden takes routes through the same path: a capability is invoked, it passes through the Governance Engine, and only then does it either execute or come back to you as a question. That path doesn't change based on how impressive or mundane the task is. Sending an email, touching a file, signing in to a site: all of it passes through governance first.
In practice, this is organized into three permission categories: Authentication, Browser Automation, and System Actions. For each, you decide the rule (Always allow, Ask Every Time, or Never) and Alden follows it consistently. Ordinary clicking and typing you already asked for doesn't interrupt you mid-task; it's the genuinely consequential sub-actions, like a file upload, a purchase, or a payment screen, that trigger a pause.
Full autonomy sounds appealing until the assistant does something you didn't actually want: an email sent too early, a form submitted with the wrong data, a purchase you meant to review first. The cost of a wrong autonomous action is rarely proportional to the time it saved. Asking first has a small, predictable cost (a moment's pause, a confirmation) in exchange for removing that risk almost entirely.
Governance-first design also means the boundary is yours to set, not fixed by us. Some people want Alden to always ask before touching email; others are comfortable letting routine tasks run without interruption but want a hard stop before anything involving money. Both are reasonable, and both are just a permission setting.
The Governance Engine is one of the most concrete parts of the Alden product precisely because it needs to be. As new skills and capabilities are added, they're built to route through the same governance path rather than getting a special exception, that consistency is the point. If you want the full breakdown of how the three permission categories work, we've written that up in the governance and permissions guide.