TJ Larkin

AI workflows

Why does ChatGPT or Claude forget my work, and how do I stop it?

Short answer

Regular chats don't reliably save things, and long chats run out of context. I lost a four-hour website build that way. For anything you're building, work in tools that read and write folders on your computer, like Claude Cowork, Claude Code or Codex, so the work lives in files instead of chat memory.

I once spent four hours building a website in a regular chat, did something else for an hour, came back, and it had restarted. I asked for my website and it gave me a totally different one. It made something up. I yelled at it for 20 minutes.

"Regular chats don't save things, or at least you can't count on them to. Use folders on your hard drive to save things."

Why does the chat lose things?

Two reasons I've run into over and over:

  1. Chat memory isn't storage. Come back a day later and it says it doesn't have it.
  2. Context runs out. One of the biggest complaints people had was that before the AI got something completely done, it would run out of context and fall apart, because it lost its understanding of what it was doing. Like Dory in Finding Nemo. Bigger context windows helped a lot, but long chats still drift.

Drift cuts the other way, too. I spent four hours planning a build in my main chat, and an hour or two in, it said "oh, we built this thing for" one of my other brands. I hadn't mentioned that brand in six months. One giant chat mixes your projects together.

What should I use instead for builds?

Tools that work out of folders on your computer. That's the biggest difference between regular chat and Claude Cowork, Claude Code or Codex: they read and write real files, so the work persists and they don't forget as much.

You don't even have to manage the folders. I don't know what's in most of mine. Claude organized them for me.

How should I split chats by purpose?

UseWhere
High-level brainstorming you don't need remembered in detailMain chat
Long builds that need continuityA folder-backed workspace (Cowork, Claude Code, Codex)
A middle groundProjects with uploaded files. More manual, but better than nothing.
Quick, disposable web tasksBrowser-use mode only

About that last row: Claude's browser-use mode loses its memory when it switches tabs. Strategy happens in regular chat. Browser use is for throwaway tasks.

What's the simplest habit that fixes this?

Plan in chat, then push the work into the coding tool so it saves everything. I treat the chat as a product manager writing instructions and the coding agent as the developer who keeps the files: plan with one AI, build with another.

And keep one AI coding tool per project. Bouncing between tools on the same project brings the same context and memory problems back.

To stop re-explaining your business in every new chat, put it in a file once: how to give AI context about your business.

Answered by TJ Larkin. From AI training sessions, client calls and Local Media HQ podcast episodes in 2026. Last updated .

Related answers

TJ Larkin / Answers / AI workflows