Email agent

Why it matters

Email is the inbox most knowledge workers spend the most time in. Most "AI for email" products today are summarizers (give me the gist of unread mail) or drafters (write a reply for me to send). Both are useful but partial.

A full email agent is something different. It reads incoming email, decides what needs a reply, what needs the human's attention, and what can be handled autonomously. It drafts replies and sends them (with approval gating for the consequential ones). It maintains the thread context. Over time it learns the user's voice, their relationships, their priorities.

The shift is from "email assistant" to "email correspondent." Some threads the user reads and replies to themselves. Some threads the agent handles end-to-end with the user copied. Some threads escalate to the user with the agent's draft pre-loaded.

The trade-offs:

• Voice authenticity. The agent must sound like the user. This is harder than it looks; subtle cadence, vocabulary, and relationship markers matter.
• Approval calibration. Auto-sending the wrong thing is worse than not auto-sending. The right setting depends on the thread, the relationship, and the stakes.
• Privacy. The email agent reads everything; the user has to be comfortable with that.

How Ajeris handles this

Ajeris's email agent reads and writes Gmail through OAuth (read + send scopes). For each incoming email, the agent decides: skip, summarize for the morning brief, draft for the user to review, or (with explicit user authorization) send directly. The agent learns the user's voice from how the user edits draft replies over time.

Approval gating is configurable per relationship and per thread type. The agent's defaults are conservative; the user loosens as trust builds. Every send writes to the audit trail.