Should AI Work Live in a Project, Folder, or Conversation?

Organize recurring AI work by duration, shared context, and saved outputs so projects, folders, and conversations each have a clear role.

Use a project to hold an ongoing goal, folders and files to preserve official material and versions, and conversations to handle one stage of exploration or execution. Important results must return to durable files. A chat history should not be the only place where the current decision exists.

You may discuss one proposal across three conversations: the first contains the background, the second has the latest figures, and the third contains the decision. A week later, even you cannot tell which version is current.

The problem is not only that AI can forget. The work has no clear home. Give each location one responsibility.

OpenAI documents what ChatGPT Projects and Codex can hold and access. The division below—project for an ongoing goal, folder for durable material, conversation for a single stage—is my working method, not a product-mandated architecture.

A project carries an ongoing goal

A project is useful when work continues across multiple conversations, shares source material, and needs the same goal or instructions: preparing for a job search, organizing a monthly reading group, or rebuilding a product website.

OpenAI describes ChatGPT Projects as workspaces that keep related chats, files, and project instructions together. Project instructions apply within that project. OpenAI: Projects in ChatGPT

A ChatGPT Project is not the same technical object as a local Codex folder or a Git repository. OpenAI also explains that Codex can work with local folders, repositories, terminals, and developer tools, while Codex history remains separate from ChatGPT history. OpenAI: ChatGPT Work and Codex

Start with one sentence:

This project exists to complete ______ over time. It does not include ______.

Folders preserve material that must remain inspectable

An answer inside a chat is convenient, but it is a poor single source of truth. Requirements, sources, confirmed decisions, outputs, and version notes should be saved in files that you know how to review, back up, and share.

A minimal structure might be:

project-name/
  sources/       original material and references
  working/       drafts still being developed
  decisions.md   confirmed decisions with dates
  outputs/       deliverables
  README.md      current goal, entry point, and instructions

The names are optional. The essential boundary is that source material is not overwritten by a draft, confirmed decisions have a stable location, and another person does not need to read ten chats to find the current version.

A conversation should handle one clear stage

Use a conversation for exploration, comparison, review, or one bounded task. When the problem changes, open a new conversation so old constraints do not become tangled with the new objective.

One project might contain conversations for:

organizing event requirements;
comparing three public venue options;
reviewing the final invitation.

Before closing a conversation, move durable results back to the project. Write approved decisions into decisions.md, save deliverables under outputs/, and record unresolved questions as next steps. “I remember the AI said something about this” is not a project record.

Use three questions to choose the location

Question	Best location
Does this work continue for weeks under one shared goal?	Project
Must this content be saved, versioned, reviewed, or delivered?	Folder or file
Is this one exploration, comparison, or local task?	Conversation

Information can begin in a conversation, return as a file, and become part of a project’s durable context. These are not mutually exclusive containers. They form a path from exploration to preserved work.

Practice: map one ongoing piece of work

Personal AI workspace map

Project name and ongoing goal:
Out-of-scope work:
Official folder location:
Where original material lives:
Where confirmed decisions live:
Where deliverables live:
Purpose of each current conversation:
What each conversation must save back:
Current-version entry point:

You are finished when you can close every conversation and still find the goal, sources, approved decisions, and current result from the folder.

The next lesson asks which instructions should remain across many tasks in a Project or AGENTS.md, and which requirements should stay temporary.