Articles · Well Crafted AI

Building thoughtful, user-centered AI experiences.

Well Crafted AI is the publication where I think out loud about agent UX, AI-Ready Design Systems, MCP, trust, and the new shape of human-in-the-loop work. Articles publish on Substack; this page is the index.

Well Crafted AI — illustrated banner showing a designer and an AI collaborator at adjacent workstations, with framework diagrams in the background

The thread across the work: good UX patterns don't expire when the underlying tech changes — they get a better engine. Most of these articles are short reads (2–4 minutes), built around a single sharp idea, with a hard close and a link back to the project work that inspired it.

Publication Well Crafted AI
Author Jay Bellew
Cadence Weekly on Substack
Latest The AI/UX Blueprint v2

All articles

All articles · most recent first
Cover image for A Name Tag for the Agent
OCT 17 · on Substack

A Name Tag for the Agent

Google's new coworker agents get their own email, calendar, and spot in the company directory. That small choice makes AI at work much easier to understand.

Read on Substack
Cover image for While You Were Away
OCT 9 · on Substack

While You Were Away

AI agents now work while you sleep. The screen you come back to may be the most important one they have.

Read on Substack
Cover image for Still Checking
OCT 1 · on Substack

Still Checking

OpenAI's new agent went quiet on stage for ten seconds. The problem wasn't speed. It was silence.

Read on Substack
Cover image for The AI/UX Blueprint v2
SEP 26 · on Substack

The AI/UX Blueprint v2

The five-step process I wrote in 2024 still works at the start and the finish. The middle was built for software that gives the same answer every time.

Read on Substack
Cover image for The Foreground Problem
SEP 22 · on Substack

The Foreground Problem

Two years ago I wrote that the best interaction is the one that never happens. Agents made it real, and revealed the other half of the job.

Read on Substack
Cover image for The Permission Layer
SEP 22 · on Substack

The Permission Layer

Agents can spend, send, and act on our behalf. The interface for authorizing and stopping them barely exists.

Read on Substack
A three-layer diagram: Layer 01 The human, Layer 02 The new UX layer (model, tools, skills, prompts, MCPs), Layer 03 The systems — with intent/response and action/result arrows running between them.
In progress · 4 MIN

The New UX Layer

What MCPs, agentic tools, and AI-native workflows are quietly building underneath your software — and why the next decade of design lives in the gap between you and the rest of your stack. First in a new series.

Want the next one in your inbox?

Subscribe to Well Crafted AI on Substack. New posts land weekly.

Subscribe on Substack