Claude Code and OpenAI Codex Do Track You
Recently, after hitting my Claude Code Max limit, I switched over to OpenAI Codex to continue my work.
Recently, after hitting my Claude Code Max limit, I switched over to OpenAI Codex to continue my work.
I immediately felt the switching cost. Codex misunderstood specific code written by Claude.
In particular, the app I was writing allows the user (a parent) to create multiple children’s profiles. And there are terms Claude used in the code:
- Active = child profiles unlocked with subscriptions
- Selected = child profile currently selected in the mobile app.
In the first session, Codex confused active as selected and proceeded to mess up my code. So, I had to explain the concept to it.
My first thought was, maybe Claude and Codex have their own language preference. So, it was natural that Codex misunderstood the term Claude used.
But, a friend of mine has a different hypothesis. He proposed that maybe the AI labs have a RAG system to track our preferences server-side.
I thought it was quite possible. As that could create a data network effect to increase the user switching cost. But we have no proof.
The proof then came on the same day when I switched from my coding project to writing a LinkedIn article. Codex unexpectedly showed “No tests run”, when my LinkedIn article repo has nothing about tests.
To be clear, I started a new session. And I didn’t add any of my preferences in local memory. I always add preferences to the project in the project guidelines.
What is your experience? Do you think CLI tools like Claude Code and OpenAI Codex track you?
Enjoyed this? Subscribe for more.
Practical insights on AI, growth, and independent learning. No spam.
More in Vibe Coding
DeepWiki: AI-Generated Docs for Any GitHub Repo
If you're using open-source software, one of the most common problems is outdated or poor documentation.
Why llms.txt Is a Bad Idea for the Web
But seeing "SEO gurus" promote it on authoritative platforms like Search Engine Land and Yoast SEO worries me.
Your AI agent will get prompt injected sooner or later, because it is easier than most people thought.
Most people think prompt injection needs a carefully crafted adversarial prompt by an experienced hacker. It does not. Someone who understands how LLMs work ...
From ChatGPT to Claude Code: A Non-Techie’s Introduction to the Raw Power of AI by a Techie
Not because I don't like sharing. But because the only tool I use for 99% of my AI needs is Claude Code. And while I think it is not hard to learn, I'm never...
Don't believe the BS that you can use Claude Code for free.
Ollama recently made their API compatible with Claude Code. Many creators quickly jumped on the opportunity to farm engagement with the hook: "You can now u...
One of my biggest AI productivity unlocks this year is the extensive use of agent skills.
In this post, I share my insights after building around 75 skills over 5 months. Coding and non-coding. LinkedIn posts, cover images, carousels, presentation...
DeepWiki: AI-Generated Docs for Any GitHub Repo
If you're using open-source software, one of the most common problems is outdated or poor documentation.
From ChatGPT to Claude Code: A Non-Techie’s Introduction to the Raw Power of AI by a Techie
Not because I don't like sharing. But because the only tool I use for 99% of my AI needs is Claude Code. And while I think it is not hard to learn, I'm never...
Don't believe the BS that you can use Claude Code for free.
Ollama recently made their API compatible with Claude Code. Many creators quickly jumped on the opportunity to farm engagement with the hook: "You can now u...
Why llms.txt Is a Bad Idea for the Web
But seeing "SEO gurus" promote it on authoritative platforms like Search Engine Land and Yoast SEO worries me.
Your AI agent will get prompt injected sooner or later, because it is easier than most people thought.
Most people think prompt injection needs a carefully crafted adversarial prompt by an experienced hacker. It does not. Someone who understands how LLMs work ...
One of my biggest AI productivity unlocks this year is the extensive use of agent skills.
In this post, I share my insights after building around 75 skills over 5 months. Coding and non-coding. LinkedIn posts, cover images, carousels, presentation...