Agentic Intake
An intake prototype where AI does the research and a person makes the decision. Requests come in through chat, and an agent assembles context by traversing Microsoft Graph.
The Problem
Legacy intake puts a queue in front of every decision. A request waits until a reviewer can research it, gather context, and assess it. Most of that time goes to research rather than judgment, so throughput is capped by the number of reviewers.
The Reframe
Analysis and research are no longer scarce labor. When AI can do the reading and preparation live, at very low cost, the division of labor flips: AI analyzes and prepares, humans judge and decide. The reviewer spends the day on the part that needs a person, the decision.
What the Prototype Does
A standalone system built around that flipped model: chat in front, an agent in the middle, a human at the end.
- Submit: a request comes in through a chat interface instead of a form.
- Research: the agent traverses Microsoft Graph to assemble the context a reviewer used to chase down by hand, then attaches its findings and a confidence score to each request.
- Queue: requests land already triaged, with the hard cases flagged for attention.
- Review: a human opens only what needs judgment, with the research already done.
- Analysis: a view over what's flowing through the system.
Status
A functional prototype that proves out the loop and the confidence UX, not yet a shipped product. It's testing two questions: how much of intake can move to "AI prepares, human decides," and whether the handoff can be made clear enough that a reviewer trusts it.