ChatGPT Projects vs. Agents: Context, Actions, and Handoffs
ChatGPT Projects give a conversation persistent instructions and files. ChatGPT agents (ChatGPT Work) take multi-step actions to complete a goal. Understanding what each one keeps, and what it loses, helps you choose the right setup for each task.
Updated 7 min read
ChatGPT Projects and ChatGPT agents (running under the ChatGPT Work name as of July 2026) solve different problems. Projects give a series of conversations a shared set of instructions, files, and a persistent thread. Agents take a goal and carry out a sequence of actions to reach it. Tools like fenbs address the gap that both leave open: the record of decisions made, work reviewed, and what the next session should start from.
The two are not competing options for the same job. Most people who use one will also use the other at different points in the same project. This guide explains what each does, what it preserves, and when to combine them.
| ChatGPT Projects | ChatGPT Work (agents) | |
|---|---|---|
| Purpose | Persistent context for a series of conversations | Multi-step task execution toward a goal |
| What persists | Instructions, uploaded files, conversation thread | Steps taken and output produced in the session |
| User control | You send each message; the assistant responds | The agent runs steps autonomously until the goal is reached or it needs input |
| External actions | None; conversations only | Web browsing, code execution, file management, connected tool actions |
| Best for | Ongoing work with a stable brief and shared files | Tasks that need a sequence of actions to complete |
| Context gap | Does not track task status or decisions across sessions | Each session starts fresh without prior project history |
What ChatGPT Projects do
A Project in ChatGPT is a container for related conversations. You write a custom instruction for the project — the role the assistant fills, how it should respond, any standing rules — and upload reference files. Every conversation inside that project starts with those instructions and files already loaded. The conversation thread persists, so you can scroll back through previous exchanges.
Projects are about context: they give the assistant the same brief every time so you do not have to repeat yourself. They do not track the status of tasks, record decisions you made, or carry a "where we left off" note from one conversation to the next. Each conversation is a fresh exchange inside a stable context container.
For more on best practices with Projects, ChatGPT Projects best practices covers how to structure instructions and files for different use cases.
What ChatGPT agents (ChatGPT Work) do
ChatGPT Work is the agentic mode where the assistant takes a goal and runs a series of steps to reach it. It can browse the web, write and execute code, manage files, and call connected tools without you prompting each step. You describe what you want, set any constraints, and review the output when it finishes.
The agent session is task-oriented, not conversation-oriented. It logs the steps it took so you can review them, but it does not carry project history from previous sessions. A Work session started tomorrow knows only what you give it at the start. If that task depends on a decision made last week, you bring that context yourself.
What carries between sessions in each
In Projects: the custom instructions and uploaded files always load, but the active task status and any decisions from past conversations are not surfaced automatically. You can scroll back through the thread, but a new conversation in the project does not know "we agreed last Tuesday to use format X."
In ChatGPT Work: the steps and output of the current session are logged. Nothing from a prior session is loaded unless you paste it in. If you run a weekly research task with Work, each run starts from zero unless you supply the previous output as input.
The gap in both cases is the same: decisions, standing rules, and task status are not stored in a form that the next session reads automatically. That is what a shared work record outside the chat is for.
Which to use for which task
Use a Project when: you have an ongoing relationship with a subject (a client, a codebase, a research area), your brief is stable, and most of your work is conversational — writing, reviewing, answering questions. Projects shine for work that is continuous and dialogue-driven.
Use ChatGPT Work when: the task has three or more distinct steps, involves reading from external sources, or produces an output that results from a series of actions rather than one response. Agentic mode is for tasks that would take you several manual steps and benefit from automation under your review.
- Project: "Review this proposal draft against our style guide and suggest edits."
- Project: "Answer questions about our product based on these documentation files."
- ChatGPT Work: "Search for the five most recent papers on this topic, summarize each, and produce a comparison table."
- ChatGPT Work: "Scan this codebase for functions matching this pattern and list the file and line number for each."
Bridge the context gap between Projects and agents
fenbs holds the project decisions, standing rules, and Where we left off that neither ChatGPT Projects nor ChatGPT Work sessions carry automatically. Connect ChatGPT from Integrations and ask it to read the project context before each session — whether that is a Project conversation or a Work run. When it finishes, it saves progress back so the next session knows where things stand.
Create a free accountFree for one person · No credit card · Works with Claude, ChatGPT, Cursor and more
Combining Projects and agents in one workflow
Projects and ChatGPT Work are not mutually exclusive. A common pattern is to run a Work session to gather and process information, then use a Project conversation to review, refine, and make decisions about what the Work session produced. The Project has the stable brief and file context; the Work session does the legwork.
The bridge between them is a handoff note: what the Work session produced, what was reviewed and accepted, and what the next step is. That note lives outside both — in a document, a notes app, or a fenbs project — and is the starting brief for the next session, whether it is a conversation or another Work run.
How to use ChatGPT Projects and agents with fenbs
fenbs holds the shared record that both modes lack on their own.
- Create a free account at fenbs and add a project for your work.
- Write the standing rules and current tasks in fenbs: what the assistant may do, what needs your review, and any constraints on output.
- When starting a ChatGPT Project conversation, paste the fenbs context summary or read it through the Integrations connection.
- When starting a ChatGPT Work session, paste the current task and its standing rules as the starting brief.
- After any session, record what was completed and the next step in fenbs so the work is ready for the next person or tool.
For a direct comparison of how ChatGPT Projects compare to Claude's equivalent, Claude Projects vs. ChatGPT Projects covers the key differences. For the agentic mode in more depth, ChatGPT agent mode goes further into how Work sessions are structured.
When to keep work outside both
Projects and ChatGPT Work are both contained within the ChatGPT platform. If your team uses multiple AI tools — Claude, Cursor, Gemini CLI — the context they each have is separate. A decision made in a ChatGPT Project is not visible to a Claude session unless you carry it there. For work that spans tools or people, a shared record outside any one AI platform is what makes that portable. fenbs does not connect to ChatGPT Work's agentic actions directly; it supplies the context and records what was decided and checked.