Chat case

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

Published 2 November 2026 · 18 min read

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.

“Will we make launch day?”: a woman in a sage linen jacket stands by a pale oak armchair on a city square, next to a stack of three rounded tiles in green, amber and blue, with a few pale-blue speech bubbles beside her

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.

WAM Agentonline
I'm Maya, head of digital at Juniper Home, an online furniture store. We're relaunching our website and this is the project topic from our team group for the last two weeks, copied as is. Give me the project status: overall, then each workstream as green, amber or red with one line why. Then the top risks to the launch date, and who owes what to whom, including who is waiting on whom. Today is Friday, 30 October.[the project topic: 44 messages, Mon 19 – Fri 30 Oct]18:33✓✓
Overall: Amber. The plan still holds on paper, but three deliverables have no owner and the payment chain lands in the same week checkout was due to finish.Workstreams- Build & checkout — Amber. Cart and address steps are done on schedule; the payment step is stuck waiting for live keys, which arrive right at Tom's 6 November finish line. - Payments & verification — Amber. Contract signed and documents uploaded 27 Oct; verification is 5–7 business days and the provider says end of next week. - Design — Amber. Checkout mockups landed a day late but are final; order-confirmation email designs are unassigned and Priya was never briefed. - Content & product data — Amber. The 23 SKU failures are closed (3 discontinued, being removed); 42 sofa photos for your Black Friday hero products are still missing and the supplier has already slipped once. - Redirects & URL mapping — Amber. Nina has old URLs but not new ones until 2 November; mapping is a full week from there, leaving no slack before the freeze. - Load testing — Red. Nobody owns it, there is no date, and it decides whether 6x Black Friday traffic holds. - Staging & QA — Amber. Staging due 9 November as planned, but no named QA owner and no test window agreed after it.Top risks to 23 November1. Load testing is unowned. Raised 28 Oct, still "let's discuss". Any failure found after the 16 Nov freeze is a Black Friday outage, not a bug fix. 2. Payment verification is the critical path. Verification completes ~6 Nov, exactly when checkout was due to be finished. No room to test payments before freeze, and a provider slip pushes checkout into it. 3. The redirect chain has zero slack. Tom's export already slipped 23 → 26 Oct. New URLs 2 Nov + one week mapping = 9 Nov. A second slip puts 1,800 pages past the freeze. 4. Sofa photos. 42 hero SKUs, supplier now saying "mid next week" after already missing one date. Placeholders on hero pages during your biggest sales week. 5. Gift cards still undecided. 4–5 dev days plus a second integration, open since 21 Oct, and Chris is asking again.Who owes what- Tom → Nina: new URL export, Monday 2 Nov. Nina is blocked on it and cannot start mapping. - Rafael → Tom: provider live keys after verification (~3–6 Nov). Tom's payment step waits on this. - Provider → Rafael: verification completion, end of next week. - Supplier → Alex: 42 sofa photos, mid next week. Alex and the hero pages wait on it. - Alex → Tom: removing the 3 discontinued SKUs from the catalogue — nobody has actually picked this up. - You → Chris: answer on gift cards, owed since 21 Oct. - You → team: a load-testing owner, open since 28 Oct. - You → Priya: decision on order-confirmation emails, hers or Tom's, asked today. - Nothing outstanding: Dana (contract returned 21 Oct), Sam (SKUs closed 29 Oct), Priya (mockups final).Want me to draft the three messages you owe — load-test owner ask, gift-card answer to Chris, and the URL-export nudge to Tom before Monday?18:33
Message

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.

WAM Agentonline
Will we make 23 November? Walk me through the critical path from today to launch with dates, tell me where the slack runs out, and give me the two decisions I need to make this week.18:34✓✓
Yes — 23 November holds, but only on two dates: 2 November and 6 November.Critical path- Fri 30 Oct. Checkout cart and address steps done. The payment step is idle waiting on verification and nobody owns load testing. Last quiet day you get. - Mon 2 Nov. Tom ships the new URL export; Nina starts mapping 1,800 pages. Sofa photos due from the supplier around 4 Nov. - 3–5 Nov. Provider verification window (5–7 business days from 27 Oct). Their own guidance is "end of next week", i.e. 6 Nov. - Fri 6 Nov. Last date for live keys. Tom's payment step starts the same day his checkout build was supposed to finish. Nina's mapping finishes 6 Nov if the export lands on 2 Nov. - Mon 9 Nov. Staging ready. Checkout, redirects and content all converge here. - 9–13 Nov. The only test window: end-to-end checkout with real payments, 1,800 redirects, load test at 6x. Five working days for all three. - Mon 16 Nov. Code freeze. - 17–20 Nov. Regression and blocker fixes only. - Mon 23 Nov. Launch. Fri 27 Nov. Black Friday, four days later with no post-launch fix cycle.Where the slack runs out- 6 November is the whole buffer. Payment verification is the critical path because it starts last and sits outside your control. Every day past 6 Nov removes a day from the test week one-for-one. - 2 November is the second pinch. Tom's export has already slipped once (23 → 26 Oct) and Nina's week of mapping has no float behind it. - 9 November for load testing. It can only start when staging exists, so it gets the same five days as everything else, fixes included. - Project-wide float is 1–2 days, all of it inside 9–13 Nov. There is no other cushion anywhere.Worth saying plainly: 23 Nov can move a day or two. 27 Nov cannot.Two decisions this week1. Who runs load testing — name them by Monday. My call: book an external load test for 9 Nov now and put one person from your side on it as owner. Wait a week and the test window is gone; you go into six-times traffic never having tested it. 2. Gift cards — no for launch. Four to five dev days plus a second integration, inside the final three weeks before a date you cannot re-test for. Put it on January's roadmap and tell Chris. He has asked twice; a clear no costs him less than silence.Three things to chase today, not decide: Tom's export for Monday, a firm verification date from Rafael's provider, the supplier's actual photo date. All three have already slipped once.Want the three messages drafted so you can send them before the weekend?18:35
Message

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.

WAM Agentonline
Write a status update for Chris, our CEO. He reads it on his phone between meetings: five or six short lines, overall status first, then what's on track, what's at risk and what I need from him. Include an answer on gift cards based on what you said above.18:35✓✓
Draft for Chris — five lines, phone-sized. Copy and send as is, or tell me what to change.Launch update — Fri 30 OctOverall: on track for 23 November. Solid, but thin — about 1–2 days of slack in the whole plan.On track: staging comes together 9 Nov, code freeze 16 Nov, and the big content push (1,800 product descriptions, 42 sofa photos) is closing on schedule.At risk: three things. The payment provider's verification clock is the critical path — live keys must be in by 6 Nov or the test window collapses. Nina's 1,800 redirects can't start until Tom exports the new URLs, which has to happen Monday. And load testing still has no owner.Gift cards: not for launch. It's a checkout change two weeks out from a relaunch — the risk isn't worth it. Recommend telling Chris it lands in the January roadmap instead.What I need from you: sign off on an external load test booked for 9 Nov — it's the one piece nobody owns and the only way we find out we can hold Black Friday traffic. If the provider drags past 6 Nov, I'll want you on the phone to them.The gift-card line reads oddly addressed to Chris, since the ask is about him.18:36
Two fixes. The update goes to Chris himself, so the gift-card line should speak to him directly. And the sofa photos are not on schedule: the supplier has slipped twice, so they belong under risks. Keep it to six short lines.18:38✓✓
Status: On track for 23 November, thin margin — 1–2 days of float, all of it inside the 9–13 November test window. Black Friday, 27 November, cannot move.On track: Development (Tom) and design (Priya) — build on plan, staging ready 9 November.At risk: Payments — provider verification runs 5–7 working days, live keys 3–6 November at the earliest.At risk: Redirects blocked until Tom exports the 1,800-URL list; load testing still has no owner; sofa photos — the supplier has slipped twice.Gift cards: no for launch — a checkout change two weeks out is not worth the risk; it moves to the January roadmap.Need from you: approve an external load test booked for 9 November.18:39
Message

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 on keeps the agent quiet in the daily flow: it replies only when someone mentions it or replies to it.
  • /trust group lets it keep your teammates' messages in the group, so the thread is there when you ask for a status.
  • /allow @username gives 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 M, the critical path 2 M and the first CEO update 1 M. The fix to the update cost less than a magnet. That is about 8 M for the whole case.

Text work on the default model costs 1 M per 20,000 tokens, so even a long project topic is a few magnets. The Free plan gives 200 M in the first month and 100 M every month after, with no card. Pro costs $13.90 a month for 400 M, 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 agent

Questions

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 M. The Free plan gives 200 M in the first month and 100 M every month after, with no card. Pro is $13.90 a month for 400 M.

Your AI agent in Telegram, ready in a minute

Open the bot, hit Start, and it walks you through the rest.

Create your AI agent

Free · no card · nothing to install · cancel in the bot

Free no card, nothing to set upCreate your AI agent