AI × Personal PublishingAI / 2026-09-26
The Day One AI Stopped

When AI Stops
My Personal Publishing Stops Too.
Lessons from Relying on a Single AI

Codex, which I use to produce articles for my personal site, suddenly started showing an error. I also have Claude Code at home. But moving the setup and working assumptions I had built up in Codex takes preparation. Having another AI available did not mean I could immediately resume writing.

A quiet illustration of one connection breaking while a line extends along an alternate route
Codex, which handled article production, stopped working.
Section 01 — The Assumptions an Outage Revealed

It Was More Than “The Chat Is Unavailable”

At first, I thought something was wrong with my setup. I got an authentication error. Starting a new session did not help, and another feature failed in the same way. After a while, I found a Codex service incident notice and realized the problem was not limited to my configuration.

That was reassuring, but it also made me notice something else: I had built my personal site’s article workflow around Codex more than I realized.

Before, if AI was unavailable, I could simply think, “I guess I can’t use it today.” Now AI is part of my personal production process, from organizing notes and outlining articles to implementing them in HTML and checking the result. If Codex stops, the workflow toward publishing an article stops too.

Section 02 — AI in My Personal Production Workflow

AI Is Deeply Involved in the Article Production Workflow

When I create an article for my personal site, I develop the idea, think it through with Codex, have it implement the result in HTML, and review the final page myself. AI is involved in the path to publication, not just in writing.

01 / Think
Develop the ideaChoose the experience and point I want to write about.
02 / Discuss
Think it through with CodexOrganize the key points and plan the structure.
03 / Build
Implement the articleAdd the article text to the site’s HTML.
04 / Review
Review it myselfCheck the content and layout, then decide whether to publish.

The more I get used to this workflow, the more Codex becomes a production environment that helps me maintain my personal site, rather than just a writing assistant. That is why an outage has a greater impact.

Having another AI available
and being able to move the work to it right away
are not the same thing.

Section 03 — The Cost of Switching

I Had Claude Code, but Switching Was Not Simple

I have Claude Code at home. When Codex stopped working, I thought, “Could I just switch to Claude Code?” Codex was not the only AI available to me.

But switching would mean providing it again with reference materials, decision criteria, current work status, folder structure, runtime environment, and working rules. I had built a fairly complex setup around Codex, so preparing the switch was heavier than I expected.

For a single article, waiting for recovery seemed faster than rebuilding the environment. I decided not to move to Claude Code this time. Having another AI on hand did not make it an immediate substitute. I only understood that when the outage happened.

Section 04 — What Single-AI Dependence Really Means

What Had Accumulated Around One AI Was the Context for My Work

I can use both Codex and Claude Code. Even so, the context for my day-to-day article production had accumulated around Codex. Unless I can move the working rules and decision criteria too, the other AI is not an immediate alternative.

Usual Setup

Continue with Codex

Produce articles using the reference materials, decision criteria, work status, and runtime environment already in place.

Switching Cost

Move to Claude Code

Provide the same context again and prepare the environment before article production can resume.

Having multiple AIs and being able to switch work between them are different things. What stopped this time was not my choice of AI, but the article workflow running inside a carefully built environment.

Section 05 — Preparing to Keep Creating

Small Preparations That Help Me Keep Publishing

That does not mean I need to keep my Codex and Claude Code environments perfectly synchronized at all times. A first step is to keep source notes, decision criteria, and work in progress somewhere I can retrieve when needed, rather than locking them inside a particular chat history.

If switching still takes time, I can decide case by case whether to move the work for one article or wait for recovery. I do not need fully automatic failover; I need to know how much of the work I can transfer. That is a small preparation for keeping my personal site going.

01 / Choose
Choose the work to transferChoose the parts of article production I cannot afford to stop, rather than moving everything.
02 / Keep
Keep working context outside the AIKeep source notes, decision criteria, working rules, and work in progress accessible.
03 / Decide
Decide whether to switch or waitCompare the effort of preparing Claude Code with the time it may take for Codex to recover.
Summary

Even when another AI is available,
the article-production workflow may not be immediately transferable.

I had Claude Code at home, but the cost of moving my article-production setup and context from Codex was high, so I waited for recovery this time. Rather than assuming multiple AIs are enough, I want to consider how much of the necessary context I can carry between them.

When I decided to write this article, Codex was experiencing an outage. I could not publish an article about personal dependence on AI because that AI was down. As an outage drill for the AI era, it was almost too well designed.