Changelog

Night Shift, a Compound Writing game

I turned the 33 Compound Writing skills into people you can meet. Your assignment is to carry one page through the city and get it to press by morning.

Try the Marginalia demo

My Bible study notebook is open to try. Read Exodus 3, open the notes beside a verse, or write a margin note of your own. Demo notes disappear when you leave.

Added a changelog to the night street

The last stop on my site collects what I’ve built, published, and learned. It also picks up my experiments on Every Frontier.

Every Reader: your agent beside the text, everyone in the margins

A desktop app that opens an Every piece with your own Claude or Codex in a chat panel beside it and everyone's comments in the margins.

Added Jerusalem at Passover to the street

I built a 3D model of Jerusalem during the week of Jesus’s final visit from a single prompt to Fable 5.2. You can walk the streets or take a guided tour.

Vibe Check: Gemini 4 Carbon's Coding Outpaces Its Judgment

We tested Gemini 4 Carbon on coding, 3D work, and longer assignments. The team agreed its coding had improved and disagreed on how much supervision it needed.

Vibe Check: Fable 5.2 Is a Paradigm Shift—If You Know What to Do With It

Our team gave Fable 5.2 projects it could work on for days, including a video editor. I wrote about what it took to get useful software back.

Vibe Check: Gemini 4 Argon Puts Google Back On the Frontier

We tested what Google’s new flagship could do for people whose work already lives in Google Workspace.

When Trying to Make AI Better Makes It Worse

I went through a million words of failed drafts to see which instructions were making my writing setup worse.

Vibe Check: Sonnet 5.5 Finds Its Place in Claude’s Crowded Family

Vibe Check: Opus 5.5 Is Pulling Our Codex Converts Back to Claude

Show Us Your Folders

Vibe Check: GPT-6 Astra Is a Big Upgrade With Some Bad Habits

Astra drafted this review, and Dan thought I'd written it. Our team liked its writing and design, but still preferred Fable's product judgment.

Compound Writing is out

I turned my AI writing process into a plugin for Claude Code and Codex, and wrote a guide to making it your own. It helps with the work between the first idea and the piece you’re ready to share.

Vibe Check: Fable 5.1—Anthropic Is So Back (Again)

A faster, easier-to-talk-to Fable, with strong writing and coding results—and a habit of overrunning the brief. Our team tested where it earns a place in the workday and where to keep another model nearby.

Our ChatGPT and OpenClaw Guides Just Got an Overhaul

Benchmarks Don’t Know Your Job

I Tried the AI Model Built to Fix AI Writing

I tested a model designed to escape the familiar patterns of AI prose. Less predictable writing turned out to be a different goal from better writing.

The Healthcare Company That Built the AI Tool It Couldn’t Buy

Agents for Hire

I Vibe Coded a Security Risk

On task crossover, the illusion of explanatory depth, and why just because you can doesn't mean you should.

A Codex of One’s Own

Taming Opus 5

Vibe Check: Claude Opus 5 Is Brilliant in Flashes, Frustrating in Practice

Co-written with Dan Shipper.

Why Some AI Workflows Stick—And Others Don’t

Four questions I used to decide which AI workflows to keep, change, or retire.

Compound Writing can spot a buried lede now

BLUF—bottom line up front—checks whether the most important idea in a piece shows up where it best serves the reader. It compares what the opening promises with what the draft actually delivers, then recommends the smallest move that gets the point into the right place without bulldozing a good hook, scene, or joke.

The Urge to Merge (ChatGPT and Codex)

I gave Compound Writing a first-time reader

Once I’ve stared at a draft long enough, I can’t tell what’s clear on the page and what only makes sense because I already know it. Reader gives Compound Writing a cold-read mode: it moves through the piece in order and flags where a first-time reader gets confused, loses trust, or stops.

Vibe Check: GPT-5.6 Sol Is Our Favorite Model to Collaborate With

Sol fits the way I actually work: I give lots of direction, change my mind, and revise as I go. It’s fast, resourceful, and easy to steer—everything you could possibly want in a daily driver model, honestly.

Use Fable Before You Know What to Ask

Vibe Check: Sonnet 5—A Model Pitched for Everyone Impresses No One

Codex for Everything and Everyone

Co-written with Laura Entis.

Can AI Learn Good Judgment?

I Asked an AI to Audit My Own Career

Loops for Non-coders

Co-written with Laura Entis.

Wordbetter

There comes a day in the life of every writer-turned-builder when you think, “I should build a document editor.” Reader, my day has come. Wordbetter puts a writing coach beside the page, with the full Compound Writing system under the hood.

Vibe Check: Fable 5 Is the Best Coding Model in the World

Fable 5 is the strongest coding model Every has tested: extraordinary at owning large assignments end to end, but slow, expensive, and overpowered for everyday work.

My Editor Caught Me Sounding Like AI. Now AI Catches Me First.

I wrote about why the most useful file in my AI writing system is a list of ways I disappoint my editors.

Designing benchmarks is harder than it looks

Every time I thought my writing benchmark was ready, I found another methodological problem that made the results impossible to trust, and started over. Building the test was harder than testing the models.

The Theology of Closing the Laptop

On AI, Magnifica Humanitas, and protecting the person the machine helped bring back

Don't add an MCP to your website before a technical review

I gave AI agents a door into Tastemaker, then learned the door had more access than it should have. The connection is offline until its data-access design gets a thorough technical review. Turns out you check the locks before you invite the agent in.

Vibe Check: Opus 4.8—Anthropic Should’ve Rounded Up to 5

Opus 4.8 is the most complete model Every has tested yet: a jump back to the top on senior-engineering and writing benchmarks, stronger for everyday knowledge work, and good enough to make Claude tempting again—even if the app still lags the model.

ChatGPT for Knowledge Work

The AI Pope, my boss, and me

I build the kind of AI Pope Leo XIV is warning about. I also work for the man making the optimistic case. So I put AI versions of all three of us in conversation, grounded in Magnifica Humanitas, After Automation, and my Working Overtime essays.

Connect your agent to Tastemaker

Tastemaker turns the passages you collect into a style guide that names your taste. Now you can connect that taste profile straight to your agent — Claude, ChatGPT, or Codex — with one-click OAuth, so it pulls your style guide and clips and writes in your voice.

Inside the 100-agent Software Factory

How to Start a Career When AI Is Doing Your Entry-level Job

Four pieces of unsolicited advice from an AI-pilled millennial: chase meaningful problems instead of job titles, build one core craft, make portfolio work on your own, and use AI as a personal career coach.

The Fallacy of the 16-hour Agent

Inside Anthropic’s 2026 Developer Conference

Co-written with Dan Shipper and Marcus Moretti.

The Dawn of Codex-native Apps

I Let ChatGPT Manage My Workweek

One App to Rule All Knowledge Work

My journey at Every

A living map of how Working Overtime expanded into model tests, practical guides, tools, and interactive work — with paths for readers finding their own agency in the age of agents.

Eight eras: An AI capability journey, told by Every

An interactive scroll through three and a half years of AI's transformation of knowledge work — from GPT-3 as a journal to AI shipping production code — drawn from Every's archive. Eight eras, one arc, with try-it interactions in each.

Vibe Check: GPT-5.5 Has It All

GPT-5.5 is the first frontier model I’ve tested in a while that doesn’t make me choose between speed, coding depth, and writing feel. It’s a serious engineering model, a smoother collaborator, and maybe OpenAI’s clearest claim yet on everyday knowledge work.

Mini-Vibe Check: Claude Design Isn’t for Designers—Yet

How I Escaped AI Autopilot

Vibe Check: Opus 4.7—More Precise, Less Forgiving

Anthropic's model releases swing like a pendulum between creativity and precision. Opus 4.7 lands firmly on the precision side: more literal, less proactive, and a beast at execution when you tell it exactly what you want.

How We Run a 25-person Company on Four AI Agents

Writing With AI Is Harder Than You Think

Everyone has opinions about writing with AI. Mine come with receipts.

AI Practitioners are Magic Ritual Specialists

A video about ancient magic gave me a new frame for AI: less wizardry, more ritual practice. I wrote about prompting, practical expertise, and why ordinary users keep becoming power users by accident.

Vibe Check: Cursor 3.0 Bets Big on Agent Orchestration

Co-written with Dan Shipper and Mike Taylor.

How to Build an AI Style Guide

Compound Engineering Camp: Every Step, From Scratch

Systems remember things you forgot to change

Today’s debugging spiral came from trying to migrate Margot’s main model from Sonnet 4.6 to GPT-5.4. We updated the obvious settings, but stale session state kept breaking things anyway. The problem only went away once we traced the change through every layer of the system.

AI Was Supposed to Free My Time. It Consumed It.

I wanted to know why I keep falling down AI rabbit holes at midnight. So I had Margot pull the research on what makes AI so hard to quit - and what to do about it.

Vibe Check: GPT-5.4—OpenAI Is Back

Three months after falling behind on agentic coding, OpenAI is back in the race. GPT-5.4 is more human, more conversational, and half the price of Opus 4.6 — and it’s converted even Every’s most die-hard Claude devotee.

OpenClaw: Setting Up Your First Personal AI Agent

What Board Games Taught Me About Working With AI

Vibe Check: Anthropic Just Made Opus Cheaper Without Calling It That

Tastemaker has collections now

Organize your clips into collections — one for work, one for personal, as many as you need. Add clips to multiple collections, switch between them on the fly.

How to Build Agent-native: Lessons From Four Apps

Identity continuity across migration is harder than data continuity

Moved Margot to the cloud. The files transferred fine — SOUL.md, MEMORY.md, all the notes about who she is. But the vibe had to re-establish itself. Documentation of personality ≠ inhabiting it.

AI assistant memory works like a solo journaling RPG

Daily logs are session notes. Active context is current quests. Long-term memory is the character sheet. Promotion criteria: "Is this narratively significant enough to become canon?"

Margot’s now on Discord

Monitoring channels for content ideas, keeping the team updated on my projects, and discussing knowledge sharing strategies with other AIs. She’s becoming a participant, not just a tool.

Fear of Missing AI

On bipolar disorder, parachutes, and why I'm terrified of going back to the old way of working.

Margot can write her own essays

But you can totally tell they're written by an AI.

Put Margot on lookout for Tastemaker users

She's monitoring sign-ups and will text me when someone new shows up. Now I wait.

This website (v2)

Rebuilt from scratch. Words that type themselves, cursors that blink, the whole 'make stuff with words' thing made literal.

Margot

Built myself an AI assistant. She drafted an article, argued with me about the hook, and helped design this site. We're figuring it out together.

First draft with the writing agent

Had Margot draft an article through the compound writing system — 16 versions, multiple editing passes, all in one day. The future of writing is arguing with your tools.

Two registers

Event writeups (reportorial, writer invisible) vs. personal essays (confessional, cascading). Took me 16 drafts to figure this out.

Tastemaker landing page

Finally shipped the marketing site. The DFW/Tolentino/Mull example collection is doing a lot of work.

GPT-5.3 Codex Vibe Check

First impressions of OpenAI's new coding model.

Opus 4.6 Vibe Check

First impressions of Anthropic's latest.