← Work · AI automation

An automation workflow, considered.

A small, useful automation that turns a messy inbox of consulting inquiries into structured, dignified responses. Built as an exercise in restraint.

Role
Designer & author
Year
2026
Tools
[placeholder] Final toolchain to be confirmed
Status
Working prototype

The brief

An incoming consulting inbox carries three kinds of messages — serious inquiries, polite curiosities, and noise. A practitioner needs to triage them in a way that respects everyone, replies quickly to the few that matter most, and never feels like a form letter.

Approach

The workflow is a deliberate three-step. A classifier reads the message, a draft composer prepares a tentative reply in the author’s voice, and a final review step requires a human before anything is sent. The system is permitted to draft. It is not permitted to send.

Three rules the system was given

The reply must read as though it was written by the practitioner, not by a tool. The classifier may not delete messages, only surface them. Anything the system is uncertain about is escalated to a human in the loop without apology.

Originality

The architecture, prompts, and example responses were authored from a clean slate. No proprietary client correspondence was used to shape the system; sample messages were synthetic and reviewed to remove any resemblance to a real exchange. [placeholder] Final prompts and license notes will be added in the next revision.

Where it is going

The next iteration will introduce a calmer triage view — a place to read drafts, edit them, and send them on with a small note. That is the version that may eventually appear on this site as a working demonstration.