Chat case
AI for project management: a project status from the team chat in a minute, with risks and who owes what

Every project has a topic in the team group where the real status lives, scattered across “Monday”, “end of next week” and “good question”. We copied two weeks of a website relaunch, 44 messages from nine people, and sent them to our test agent. In 27 seconds it returned the status by workstream, the risks to the launch date and who owes what to whom. Two more messages gave the critical path and a six-line update for the CEO. Below are the real replies, the cost and the requests to copy.

The short answer
Copy the project topic from your team group, send it to your AI agent and write one line: “Give me the project status: overall, each workstream as green, amber or red with one line why, the top risks to the launch date, and who owes what to whom.” Add today's date and say who you are. In our test the status came back in 27 seconds.
Then keep asking in the same chat. “Will we make the date?” gets you the critical path with dates and the place where the slack runs out. “Write an update for the CEO” gets you six lines he can read between meetings.
That is what people usually want from AI for project management: not another board to fill in, but someone who reads the whole thread, connects the dates and says plainly what is late and who is waiting on whom. The agent starts in Telegram and works in DMs, groups and forum topics, so one topic per project is a natural fit.
The test: two weeks in the topic of a website relaunch
We wrote the project ourselves. Juniper Home is a made-up online furniture store relaunching its website. The new site goes live on Monday 23 November, four days before Black Friday, and code freeze is on 16 November. Maya, head of digital, runs the project. In the project topic with her: Tom, the dev lead; Priya, the designer; Nina, who handles SEO and redirects; Alex, content and product pages; Rafael from finance, who owns the payment provider; Dana from legal; Sam from the warehouse; and Chris, the CEO, who drops in now and then.
Two weeks, Monday 19 to Friday 30 October, 44 messages. It reads like a real project topic. Plans are announced on day one and then quietly move. One person waits on another, who waits on a third. A supplier promises photos “end of next week”, then “mid next week”. The CEO asks for a new feature twice. Someone asks “who is doing load testing?” and gets “good question”. Two people joke about sofa samples.
A few messages from the topic, so you can see what the agent had to work with:
We copied the whole topic and sent it to our test agent on 30 October 2026, in one message with the request on top. The company and the people are invented. The agent's answers are real and quoted as they came.
Step 1. Two weeks of the topic in, a project status out
The request asks for four things at once: one overall verdict, a colour per workstream, risks tied to the launch date and the chain of who owes what. Today's date matters again: it turns “Monday” and “end of next week” into real days.
Twenty-seven seconds after the message landed, Maya had the status. We checked it against the topic line by line: every date, every name and every slip in it is in the thread.
Three things in this reply would take a project manager an hour to put together by hand.
The colours are argued, not guessed. Six workstreams are amber and one is red, and each colour comes with the reason in one line. Load testing is red because nobody owns it and it decides whether Black Friday traffic holds. The redirects are amber, although nothing has gone wrong there yet: the agent counted that Nina gets the new URLs on 2 November, needs a week, and that leaves no slack before the freeze.
“Who owes what” is a chain, not a list. The provider owes Rafael the verification, Rafael owes Tom the live keys, and Tom's payment step waits on both. The supplier owes Alex the photos, and the hero pages wait on Alex. A plain task list would show “payment step: in progress”. The chain shows why it is in progress and whose phone to ring.
Maya's own debts are in the list too. An answer to Chris on gift cards, owed since 21 October. A load-testing owner, open since 28 October. A decision for Priya about the order confirmation emails, asked that morning. In a status report written by the project manager, these lines usually go missing. The agent reads the topic without that blind spot.
It also kept the noise out. The jokes about the comfortable sofa samples did not turn into a workstream.
Step 2. “Will we make 23 November?”
A colour per workstream tells you where things stand. The question a sponsor actually asks is simpler: will we make the date? We asked it in the same chat.
Twenty-four seconds. The answer starts with yes and then says exactly what the yes depends on: two dates, 2 and 6 November.
This is the critical path, written for people who never draw Gantt charts. The agent laid the dates from the topic end to end and found where they collide. Payment verification finishes around 6 November, the day the checkout build was meant to be finished. Staging, redirects and payments all arrive on 9 November. That leaves one week, 9 to 13 November, for testing payments, 1,800 redirects and a load test before the freeze. The whole project has one or two days of float, and all of it sits inside that week.
Then it gave Maya two decisions with a recommendation for each, and that is the part a manager can act on today. Load testing: name an owner by Monday and book an external test for 9 November. Gift cards: no for launch, move to January, and tell Chris clearly, because he has asked twice. One line from the answer is worth putting on the wall: “23 Nov can move a day or two. 27 Nov cannot.”
The agent also split decisions from chasing. Tom's export, the provider's date and the supplier's photos are not decisions. They are three calls to make before the weekend, and all three have already slipped once.
Step 3. An update for the CEO, six lines on a phone
Chris reads updates between meetings. Maya asked for one he can read on his phone, with gift cards answered.
The first draft came in 30 seconds, and the agent itself flagged one line: the gift-card answer was written about Chris rather than to him. Maya also spotted that the sofa photos had landed under “on track” in the draft, although the supplier had already moved the date twice. One message with two fixes, and 39 seconds later the update was right.
The final version does what a CEO update should. The verdict comes first, with the margin in days. Two lines of risk name the cause, not just the symptom. Gift cards get a clear no with a reason and a new date. The ask is one sentence: approve an external load test booked for 9 November. Chris can answer it with a thumbs-up.
A status is not a task list
If you need every task from a busy group with an owner and a deadline, that is a different job, and we tested it separately: AI for tasks: turn a week of team chat into a list of tasks with owners and deadlines. A task list answers “what does each person have to do?”.
A project status answers other questions:
- Will we make the date, and what does it depend on?
- Which workstream is the bottleneck, and why?
- Who is waiting on whom, and how long has the chain been stuck?
- What does the sponsor need to decide this week?
That is why the request in step 1 asks for colours, risks to the date and “who owes what to whom” rather than a list. The thread is the same; the reading is different. If the project was discussed on a call rather than in the topic, send the recording first: AI meeting minutes: send the recording of a team call, get decisions, owners and deadlines. Then ask for the status from the minutes and the topic together.
Running it inside the project topic
In our test we copied the topic by hand and pasted it into a chat with the agent. It takes a few seconds and works with any group, including one the agent is not part of.
The next step is to put the agent in the group as one more participant. The agent walks you through it: create a group, add your agent, show chat history to new members, turn on Topics, and make the agent an admin who can manage topics. In a group with forum topics, each topic is a separate session with its own memory. One topic per project means the relaunch and the spring catalogue never end up in the same status.
A few commands from the Guide cover the rest:
/mention onkeeps the agent quiet in the daily flow: it replies only when someone mentions it or replies to it./trust grouplets it keep your teammates' messages in the group, so the thread is there when you ask for a status./allow @usernamegives a teammate access to the agent in a topic, so Tom can ask “who is waiting on me?” himself.
Then the request is the same as in step 1, written in the topic with a mention. Pro adds recurring reminders and scheduled tasks, for example a status from the project topic every Friday at 16:00, so the week's report is waiting before the weekly call.
What two weeks of the topic turned into
Four messages and about two minutes of waiting gave Maya:
- an overall verdict and seven workstreams with a colour and a reason each;
- five risks to the launch date;
- a chain of who owes what to whom, her own open items included;
- the critical path with dates, the float in days and two decisions with a recommendation;
- a six-line update for the CEO, gift cards answered, one clear ask.
What to avoid when you ask for a project status
Each of these is on the side of the person asking, and each one is easy to avoid.
Sending only the last few days. The status depends on what was promised at the kick-off. In our topic, the dates from 19 October are what made the 2 November export a risk. Copy the topic from the start of the phase.
Leaving out who you are. “I'm Maya, head of digital” lets the agent put your own open items in the list and write the CEO update in your voice. Without it, you get a neutral summary for nobody in particular.
Asking for a summary instead of a status. “Summarise the topic” returns a retelling. “Give me the status, colours, risks to the date, who owes what” returns a decision tool. The words in the request shape the answer.
Sending the update without one read. The agent works from the thread, and the thread is sometimes more optimistic than reality. In our test, one line in the first draft put the sofa photos under “on track”, and a one-sentence reply fixed it. Read the update once before it goes to the CEO.
Asking “are we on track?” without a date. Give the date and what cannot move. “Black Friday is the 27th” is what let the agent say that the launch can move a day or two and Black Friday cannot.
What it costs
We read the balance in the agent's status line after each message. The project status cost 4 , the critical path 2 and the first CEO update 1 . The fix to the update cost less than a magnet. That is about 8 for the whole case.
Text work on the default model costs 1 per 20,000 tokens, so even a long project topic is a few magnets. The Free plan gives 200 in the first month and 100 every month after, with no card. Pro costs $13.90 a month for 400 , or $1.39 for the first month if you subscribe within 7 days of signing up, and adds recurring reminders, scheduled tasks and top-ups. All rates are on the Pricing page.
Your messages stay in your agent's isolated container. To answer, the agent sends requests to the model: on the default model they go through DeepSeek's infrastructure in the People's Republic of China, which you confirm during onboarding, and on Pro you can attach your own Claude or Codex, so requests go to that provider under your own account.
When the quarter closes and the owner wants slides rather than a message, the same chat does that too: AI for presentations: send a spreadsheet, get a quarterly report deck for the owner.
Messages worth copying
I'm [name], [role]. Here is our project topic for the last [two weeks]. Give me the project status: overall, each workstream as green, amber or red with one line why, the top risks to [launch date], and who owes what to whom. Today is [date].
Will we make [date]? Walk me through the critical path with dates, tell me where the slack runs out, and give me the two decisions I need to make this week.
Write a status update for [name, role]. Five or six short lines for a phone: overall first, then on track, at risk, and what I need from them.
Who in this project is waiting on whom right now, and since when?
Create your agent, copy your project topic and ask whether you will make the date.
Try it with your own agent
Free, no card. The agent starts working in the chat straight away.
Create your AI agentQuestions
Can an AI write a project status report from a team chat?
Yes. Copy the project topic, send it to the agent and ask for the overall status, a colour per workstream, the risks to the launch date and who owes what to whom. In our test two weeks of a website relaunch, 44 messages from nine people, became a status report in 27 seconds, with a reason next to every colour.
Will it tell me honestly if we are going to miss the date?
It says what the date depends on. Asked “Will we make 23 November?”, the agent answered yes and then named the two dates the yes rests on, worked out that the whole project had one or two days of float, and said plainly that the launch could move a day or two but Black Friday could not.
How is this different from pulling tasks out of a chat?
A task list says what each person has to do. A project status says whether the date holds, which workstream is the bottleneck and who is waiting on whom. The thread can be the same; the request is different. For task lists we have a separate test with a bakery chain's ops group.
Can the agent sit in the project topic itself?
Yes. Add it to a group with forum topics: each topic is a separate session with its own memory, so every project gets its own status. With /mention on it replies only when mentioned, and /allow @username lets a teammate ask it questions in the topic.
Can it send a status every Friday on its own?
Pro adds recurring reminders and scheduled tasks, so a status from the project topic can arrive every Friday before the weekly call. One-off reminders work on every plan, Free included.
How much does it cost?
Our whole case, the status, the critical path and the CEO update with one fix, cost about 8 . The Free plan gives 200 in the first month and 100 every month after, with no card. Pro is $13.90 a month for 400 .


