How to Make Claude, ChatGPT, Codex and Grok Work on the Same Project

Let me save you some searching: your AI tools do not share memory, and no product on earth honestly fixes that. Claude doesn't know what you told ChatGPT. Codex can't see your Grok conversation. Every tool wakes up with amnesia about the others, and every "universal AI memory" product you'll find is either overpromising or quietly just doing what I'm about to show you for free.

I run four AI tools on the same projects every single day: Claude, Codex, ChatGPT, and Grok. For months they stepped on each other. Overwriting work. Forgetting context. Me re-explaining my own business five times a day like a broken record. The fix that finally worked is embarrassingly boring, and that's exactly why it works.

Stop trying to share memory. Share files.

The whole method is one idea: one folder on your computer becomes the single source of truth for your work. Plain Markdown files. Your context (who you are, what you're building, how you talk), your projects (one file each: the goal, the state, the decisions), and your task board (what needs doing, who owns it, what's done).

Why plain files? Because they're the one format every AI can work with, today and next year, no matter which company wins. No proprietary memory feature. No platform lock-in. If a better model ships next month, you point it at the same folder and it's instantly up to speed.

Two kinds of AI tools, two kinds of access

Here's the distinction most people miss. AI tools split into two camps:

• File-capable tools (Claude Desktop, Claude Code, Codex, Cursor): these can genuinely open your folder, read your context, and write their finished work back as files. Give them the folder once, and they stop asking you to repeat yourself.
• Browser tools (ChatGPT and Grok in a tab): these cannot see your computer, whatever anyone claims. The fix is a handoff brief: a short plain-text block you paste in that carries who you are, the project, the exact task, and what the finished work should look like. Ten seconds of pasting, full context.

That's the honest version of "making your AIs work together." The file-capable tools read the folder directly. The browser tools get a briefing document. Both kinds return their work to the same place, so nothing gets lost in a chat history you'll never scroll back through.

The three rules that keep them from stepping on each other

Sharing files gets your tools on the same page. These rules keep them from wrecking it:

• Every task has exactly one owner. You, or one specific AI. Two tools working the same task is how work gets overwritten.
• Finished AI work lands in Review, not in Done. The AI saves its output and marks the task ready for your eyes. Nothing ships without a human click. You'd be amazed how much bad AI work this one rule catches.
• The folder is the truth, not the chat. If a decision matters, it gets written to a file. If it only lives in a conversation, it doesn't exist tomorrow.

What a day actually looks like

Morning: I open one dashboard that reads the task board file. It hands me my next action. I fire off the agent tasks: Codex gets one in its app, Claude gets one in its app, each pointed at the same folder. If something belongs to ChatGPT, I generate its handoff brief and paste it in. Midday: finished work has come back as files, with tasks sitting in Review. I read the actual output, approve or send back. The work is where it belongs, and nobody re-explained anything to anybody.

I built this living with chronic pain, so the system also asks what kind of day my body is having before it tells me what to do. But that's my version. The bones (folder, owners, review gate, briefs) work for anyone juggling more than one AI.