An inbox triage agent that earns your trust slowly
An agent that reads your inbox, suggests actions, and starts taking them only after you green-light a category enough times.
The Problem
Existing AI inbox tools are all-or-nothing — either they auto-archive aggressively and miss things, or they stay in suggestion mode forever and never actually save you time.
The Solution
An agent that watches your behavior, groups emails into 'patterns' (e.g. newsletters you skim, vendors you always reply to with the same template, calendar invites from internal people), and asks for permission to automate per-pattern. After 5 confirmations, it acts on its own.
The Opportunity
Trust-graduated automation is a generally useful pattern — apply it to email, then calendar, then Slack. Most agent products skip the trust step and that's why they fail.
Why this is interesting
The hard part of email automation isn’t the LLM — it’s the trust calibration. If the agent earns trust per-pattern instead of in bulk, the user always feels in control.
What a “pattern” looks like
A pattern is { matcher, suggested_action, confidence }:
- matcher: from-domain, subject-keyword, thread-history
- suggested_action: archive, label, auto-reply, draft-reply
- confidence: rises with each user confirmation, decays with each rejection