ChatGPT for Project Management: What It Does Well and Where It Stops

ChatGPT helps a project manager in four places, each documented by OpenAI today. Projects keep one piece of work’s chats, files and instructions together. Plugins, which carry apps built on MCP servers, let it read from and act in the tools you already use. Work takes on multi-step jobs and comes back with a finished deliverable. Memory carries useful context from one chat to the next. It is strong at drafting: plans, status updates, risk lists and meeting summaries. What it does not give you is shared state. Nothing in ChatGPT says, for the whole team, what is being worked on, by whom, and what is done, and nothing records who changed what. For that you connect a board, which ChatGPT can do through a developer-mode app.

Projects: one home per piece of work

OpenAI’s guide to projects and chats↗ (opens in a new tab) says a project keeps chats, files, instructions and sources together, and gives the test for when to make one: when the work will continue over time, produce more than one output, or depend on the same files and sources. Project instructions apply across its chats, and sources, meaning uploaded files and connected context, are available to every chat in it.

• Files that earn their place: the brief or charter, the current plan, the stakeholder list with who decides what, the last few status reports, and the statement of work.
• Files that go stale: exported task lists and spreadsheets of who is doing what. They are out of date the day after you upload them.
• Instructions: the audience for updates, the length, the date style, and what to do when a fact is missing.

You support the project manager for the Website Relaunch project.
Status updates go to the client's marketing director: plain English,
no internal jargon, 200 words at most.
Dates are US style (October 14, 2026).
Never invent a date, owner or number. If the files do not say,
write "unknown" and list it under Questions.
When I paste meeting notes, sort them into decisions, actions (one
owner each) and open questions before doing anything else.

How to scope a project, which files to keep and which memory setting to choose are covered in ChatGPT Projects best practices, so they are not repeated here.

Plugins and apps: reaching the tools you already use

ChatGPT’s connectors are now apps, and apps usually arrive inside plugins. OpenAI’s page on plugins↗ (opens in a new tab) says plugins bundle capabilities into reusable workflows and can include skills and MCP servers; you install one from the Plugins tab and sign in when asked. Each MCP server still requires its own sign-in and access, so ChatGPT reaches only what your account in that service can reach.

For a project manager the useful pattern is read freely, write with approval: let ChatGPT search the shared drive, the channel and the tracker, and keep anything that sends a message or changes a record behind a confirmation. What apps can reach, and what an admin controls, is in ChatGPT connectors and apps.

Work: the job agent mode used to do

Many guides still describe agent mode. In OpenAI’s current documentation the multi-step job belongs to ChatGPT Work. The guide to getting started with Work↗ (opens in a new tab) describes it as a way to delegate real work: use Chat for answers and quick drafts, and choose Work when you need a completed deliverable such as a brief, a deck, an analysis or a recurring update. In the desktop app a Cloud option keeps a task running after you close the app, and scheduled tasks can repeat on a cadence, such as a weekly dashboard refresh.

Good project management jobs for Work: turn a month of meeting notes into a decisions and actions summary, build a stakeholder deck from the last four status reports, or compare a vendor’s invoices against the agreed milestones and list the differences. What changed from agent mode, with twelve example tasks and the checkpoint in each, is in ChatGPT agent mode.

Memory: helpful recall, not a rulebook

Memory lets ChatGPT carry context from earlier chats into later ones, and you manage it under Settings, then Personalization. OpenAI’s page on memories↗ (opens in a new tab) gives the right warning for project work: treat memories “as a helpful recall layer, not as the only source for rules that must always apply.”

That matters for teams. Your memory is yours: a teammate’s ChatGPT does not know what yours remembers about the client’s preferences or last week’s decision. Anything the whole team, and every assistant, must follow belongs somewhere written and shared. How project memory settings interact is covered in ChatGPT project memory.

Drafting plans and status updates

Drafting is where ChatGPT saves the most time, as long as the facts come from something real. Give it the source, the audience and the length, and forbid it from filling gaps.

Using the tasks and notes below, draft this week's status update
for the client.
- Three sections: Done this week, Next week, Needs a decision.
- Under 200 words. Plain English.
- Every item must come from the material below. If a date or owner
  is missing, write "not yet set" rather than guessing.
- End with a list of anything you left out and why.

[paste the board, the notes, or both]

The same shape drafts a plan from a brief, a risk list from a plan, or an agenda from open questions. A library of these, grouped by planning, triage, status, risks, meetings, retros and stakeholder updates, is in prompts for project management with AI.

Where ChatGPT alone stops

• No shared board state. A project holds files and instructions, not a live list of work. Nobody on the team can open ChatGPT and see what is in progress.
• Your chats are yours. What you and ChatGPT worked out stays in your chat and your memory unless you share it.
• Status is only as fresh as the input. An update drafted from last week’s export is last week’s status, written confidently.
• No audit. If ChatGPT only suggests a task in chat, nothing records it. If it acts through an app, the service usually records the change under your account, so you cannot tell ChatGPT’s changes from your own.

None of this is a flaw. ChatGPT is an assistant, not a system of record. The fix is to keep the work somewhere the whole team, and every assistant, reads and writes, and where changes are recorded by name.

Connecting a board through a developer-mode app

When a board has a remote MCP server but no listing in ChatGPT, you add it yourself in developer mode. OpenAI’s developer mode guide↗ (opens in a new tab) describes it as full MCP client support for read and write tools, on the web, on the plans OpenAI lists there, and calls it powerful but dangerous: watch for prompt injection, mistaken writes and malicious servers. Write actions ask for confirmation by default. The steps in OpenAI’s guide to connecting an MCP server to ChatGPT↗ (opens in a new tab):

1. In ChatGPT, open Settings, then Security and login, and turn on Developer mode.
2. Go to Plugins and select the plus button to create a developer-mode app.
3. Enter a name and description, and the server URL including its /mcp path. For fenbs that is https://fenbs.ai/api/mcp, with OAuth.
4. Create the connection, then review the tools ChatGPT discovered.
5. Start a new chat, choose Developer mode from the plus menu and select the app.

With fenbs, the connection opens fenbs in your browser: you sign in, tick what the assistant may do, and approve. Nothing is copied or pasted. ChatGPT gets an access token that lasts an hour and refreshes, and you can revoke it in Settings at any time. On a company workspace, admins decide which plugins are available, so if an option is missing, ask yours.

What ChatGPT can do on fenbs

fenbs is a task board where people and AI assistants are members with roles. Once connected, ChatGPT starts with fenbs_whoami and fenbs_get_context, which gives it the team’s rules from the Decisions and rules page first, then the AI context notes every assistant reads. It files features, enhancements and bugs with fenbs_create_item, each with a note for the problem, a plan, and a priority from 1 to 10 where 1 is the most urgent, and moves them through To Do, Next Up, In Progress and Completed. Every change appears in History under ChatGPT’s own name, separate from yours. fenbs has no due dates, sprints or assignee field, so dates and owners go in the task note.