Is a project in ChatGPT or Claude an AI agent?
A project in ChatGPT or Claude is not automatically a full AI agent. It is primarily a workspace where you can bring together instructions, files and conversations. With a clear purpose, defined steps and a consistent output format, however, the project can work as a reusable AI assistant – and as a first step towards a more advanced agent.
The short answer
An AI agent is not a physical app or a special machine. It is an AI-based system that works towards a goal and can complete one or more tasks according to its instructions.
A project is the workspace itself. It becomes more agent-like when you give it:
- a clear goal,
- reusable instructions,
- approved knowledge sources,
- a defined way of working,
- a consistent output format,
- rules for uncertainty and human review.
The term reusable AI assistant is therefore a good description of a project that is still started and reviewed by a person. AI agent is more precise when the system can also choose its next step, use tools or complete several stages towards a goal.
From a regular chat to an automated workflow
- 1One-off prompt. You ask a question and receive an answer.
- 2Saved prompt template. You reuse the same instruction but start the work manually.
- 3Project with instructions and knowledge files. You bring the task, rules, examples and sources together in a reusable workspace.
- 4Agent with tools and multiple steps. The AI can plan parts of the work, use available tools and choose between several steps.
- 5Automated workflow. The system can be triggered by an event or schedule and complete tasks with limited human involvement.
For most organisations, level three is a sensible place to start. A person starts the work, reviews the result and remains responsible for the decision.
How Projects work in ChatGPT
A ChatGPT Project brings together related chats, instructions, files, sources and saved work. ChatGPT can use the project's content as context for ongoing work. Memory behaviour depends on the project's memory setting, your plan and your organisation's policies.
To create one, select New project in the sidebar, then add relevant instructions and files.
ChatGPT Projects can be useful for:
- reviewing documents against the same quality criteria,
- producing recurring report drafts,
- collecting research for longer assignments,
- applying a consistent tone of voice,
- keeping supporting material and decisions together for a team.
Projects can be shared, but available features and permissions vary by plan and workspace.
Source: OpenAI – Projects in ChatGPT
What happened to custom GPTs?
Custom GPTs are tailored versions of ChatGPT with their own instructions, knowledge files and selected capabilities. They have often been used as reusable assistants for specific tasks.
When this guide was fact-checked on 18 September 2026, OpenAI stated that new GPT creation and publishing were no longer available on personal ChatGPT accounts and that custom GPTs were planned for retirement. Existing GPTs could still be used during the transition. Timing and availability may differ between accounts and workspaces.
For a new personal workflow, we therefore recommend starting with a ChatGPT Project. In workspaces where Plugins are available, they can package reusable instructions together with connected tools and apps.
Sources: OpenAI – GPTs in ChatGPT and OpenAI – Plugins in ChatGPT and Codex
How Projects work in Claude
Claude Projects are self-contained workspaces with their own project instructions, chats, documents and knowledge. You can find Projects in Claude's sidebar. When creating a project, you can add documents and instructions for Claude to use in the project's conversations.
Claude Projects can be useful for:
- analysing a large collection of documents,
- following an organisation's language or quality rules,
- working on a longer writing or analysis project,
- reusing the same knowledge base across several tasks.
Anthropic is also rolling out a new version of Projects in stages, beginning with Claude Code. It can split work into parallel threads, among other features, but access varies by plan and user. Always base your workflow on the features actually available in your account.
Source: Anthropic – What are projects?
ChatGPT Project or Claude Project?
| Question | ChatGPT Project | Claude Project |
|---|---|---|
| What is it? | A workspace for chats, instructions, files and recurring work. | A self-contained workspace with chats, instructions and project knowledge. |
| Useful for | Ongoing work, research, writing, planning and reusable routines. | Document-heavy work, analysis, writing and reusing project knowledge. |
| Memory and context | Controlled by the project's memory mode, plan and workspace settings. | Uses project knowledge and instructions; newer project capabilities are being rolled out gradually. |
| Sharing | Availability depends on your plan and workspace policies. | Sharing and collaboration depend on your plan and organisation settings. |
| Is the project an agent? | Not by itself. It can work as a reusable assistant and become more agent-like with tools and multi-step work. | Not by itself. It can work as a reusable assistant and gain more agentic capabilities in advanced project modes. |
What is it?
ChatGPT Project: A workspace for chats, instructions, files and recurring work.
Claude Project: A self-contained workspace with chats, instructions and project knowledge.
Useful for
ChatGPT Project: Ongoing work, research, writing, planning and reusable routines.
Claude Project: Document-heavy work, analysis, writing and reusing project knowledge.
Memory and context
ChatGPT Project: Controlled by the project's memory mode, plan and workspace settings.
Claude Project: Uses project knowledge and instructions; newer project capabilities are being rolled out gradually.
Sharing
ChatGPT Project: Availability depends on your plan and workspace policies.
Claude Project: Sharing and collaboration depend on your plan and organisation settings.
Is the project an agent?
ChatGPT Project: Not by itself. It can work as a reusable assistant and become more agent-like with tools and multi-step work.
Claude Project: Not by itself. It can work as a reusable assistant and gain more agentic capabilities in advanced project modes.
Do not choose a tool because its name sounds more advanced. Choose according to your task, your organisation's policies, the quality of your own tests and the features actually available in your account.
How to build a simple work assistant
The same basic method works in ChatGPT and Claude:
- 1Choose a clear task. Start with something recurring, limited in scope and easy to review.
- 2Describe the purpose. Explain which problem the assistant should solve and who the result is for.
- 3Define the approved input. Specify which files and sources may be used.
- 4Describe the work steps. State the order of the work and when the assistant should ask a follow-up question.
- 5Define the output format. Specify the headings, tables or categories the result must contain.
- 6Set boundaries. Explain what the assistant must not do and how it should handle uncertainty.
- 7Test and improve. Try a normal case, an ambiguous case, missing information and conflicting sources.
Example: a quality reviewer
Review the text against the uploaded quality criteria. Separate clear errors, improvement suggestions and points you are uncertain about. Refer to the relevant criterion and do not invent information that is missing from the source material.
This does not make the AI infallible. It does make the work more consistent, easier to test and easier to improve.
Keep a person responsible
Even a well-designed assistant can misunderstand an instruction, use the wrong source, miss important information or produce a convincing but incorrect answer.
Pause before using sensitive information. Check your organisation's rules and whether the information may be used. Remove or de-identify anything that is not necessary.
Results that affect people, finances, legal matters, safety or important business decisions must always be reviewed by a suitably qualified person.
In summary
A project in ChatGPT or Claude is best described as a reusable AI workspace. With clear instructions, approved sources, defined steps and human review, it can function as a simple work assistant.
When the system can also use tools, choose between several steps and act more independently, it moves closer to what is more precisely called an AI agent.
Sources
- OpenAI – Projects in ChatGPT
- OpenAI – GPTs in ChatGPT
- OpenAI – Plugins in ChatGPT and Codex
- Anthropic – What are projects?
Last fact-checked: 18 September 2026. Features and availability may change depending on your plan, account and organisation settings.