Tuesday, 9:40. A new task appears in Task Manager: “Prepare a proposal for Nord”, due Friday and assigned to you. You read the three-sentence brief, then spend fifteen minutes searching old emails for anything that could help you get started. Instead of searching, you can ask the AI Assistant, which has your tasks, conversations and notes in front of it - here are three prompts to get you started.
Why the first half-hour costs the most
Before you write the first sentence of the proposal, you need to reconstruct the context: what Nord bought before, who spoke to them and what price we quoted a similar company in May. That knowledge exists within the company - scattered across chats, emails and the minds of three people.
AI models are rapidly entering the workplace: according to Microsoft's 2024 Work Trend Index (a study from February-March 2024), 75% of knowledge workers use AI at work, and 78% of them bring their own AI tools to work, usually without their company's knowledge or guidance. The problem is that a general-purpose chat in a browser does not know who Nord is. You can paste in the brief, but you cannot paste in the history of conversations with the client because you would have to find it first.
In Task Manager, the Assistant sits in the same window as the task and sees what you see - within your permissions. You do not need to paste anything in. You just ask a question.
How the Assistant knows what Nord is about
The quality of the response depends on what the Assistant has “in front of it”. There are three layers:
- The thread where you ask the question. In the task window, click the “AI Assistant” button below the reply field. The editor changes colour, and only you can see the question and answer - they remain in the thread with a “FOR YOU ONLY” label. The Assistant then knows the brief, the entire conversation and the attachments (Understanding context).
- Sources you select. In the AI Assistant window, the “Sources” button lets you add several items at once: the “Sales” chat, the “2026 Pricing” note, a folder of proposals and the week's calendar. If you select more than the model can process, the Assistant tells you exactly what it has not taken into account.
- Company Knowledge. The application builds a knowledge base from summaries of tasks and chats, as well as the contents of notes and files. When asked “what did we agree with Nord in June”, it finds a thread from three months ago, even if you did not select it manually (Company Knowledge).
Your permissions set the boundary: a regular user gets answers based on what they can see in the application, while an administrator has access to all tasks and chats. A colleague's private note, files in their private folder and their mail will not be included in the response, even for an administrator. See the FAQ at the end of this article to learn where data goes when you ask a question.
This is what it looks like in practice: the “Proposal for Nord” thread with three messages from Kasia Wójcik and Ola Nowak dated 19 August, followed by the editor in Assistant mode. Ola is only just entering her question, but the Assistant already has the task details, the conversation about the partner discount and the “by Friday” decision deadline in front of it.
Scenario 1: “Create a plan for me” - from a blank page to five steps
Back to Tuesday. Instead of searching through emails, you open the task, switch on Assistant mode and type:
“Put together a plan for preparing this proposal by Friday. Include what we agreed with Nord in June and our pricing from the Pricing 2026 note.”
The response you can expect (an example created for this article, not a record of a real session):
Based on the brief, the Sales chat from 12 June and the Pricing 2026 note:
- Wednesday morning - confirm the scope with Nord: in June, they asked for an option including implementation, but the brief does not mention it.
- Wednesday - prepare the quote using the rates from the note; in June, Marta promised them a 10% discount if they made a decision by the end of September.
- Thursday - draft the proposal using the template from the Proposals 2026 folder.
- Thursday 15:00 - Marta checks the quote (she attended the meeting in June).
- Friday by 12:00 - send it.
I could not find a proposal validity period in the sources - this should be confirmed.
The plan includes facts that are not in the brief - the discount and the person who promised it - because the Assistant checked the chat and the note. It does this automatically in the thread, using a knowledge base refreshed once a day. If you want to make sure it has considered a specific note, open the AI Assistant window and select it using the “Sources” button. The final sentence tells you what it does not know instead of inventing a validity period. Treat it as a draft to review, not a decision.
Scenario 2: “Turn this into a checklist” - so the plan does not get left in the conversation
“Turn this into a checklist.”
The Assistant returns the same points as a list with tick boxes. Click an item when it is done; click it again to undo. A list created in a private conversation is yours - your personal plan for the day.
It works the other way round too: when a task has a long description but no plan yet, simply say, “Turn the task description into a checklist I can tick off.” In the screenshot below, Ola asks for this in the “Website project” task and gets five items - from “Agree the structure of 5 subpages with client ABC” (already ticked off) to “Upload the content in both languages and complete the website handover with the client” - with a “1 of 5 done” progress bar.
When the whole team needs to see the progress, move the items to the task description as a to-do list. The list can be ticked off directly in the task card, the progress bar beneath it fills up, and every completed item shows the name of the person who ticked it off. Marta ticks off “quote checked” at 15:20 on Thursday, and no one has to ask in the chat whether it is done. Learn more: Checklists.
Scenario 3: “Write the first draft of an email” - and check it before sending
It is Wednesday and the quote is ready, but before drafting the proposal, you want to confirm the scope with Nord. You write in the same thread:
“Send an email to piotr.nowak@nord.pl asking whether the proposal should include the implementation we discussed in June. Keep the tone factual and do not mention prices.”
The Assistant suggests something like:
Subject: Proposal - scope confirmation Dear Mr Nowak, We are preparing the proposal we discussed on 12 June. Before finalising it, could you confirm whether it should also include on-site implementation (2 days), or only the licence and remote configuration? We will send the proposal by Friday.
Task Manager does not send the email automatically. You get a “Message ready to send” card with “Send” and “Cancel” buttons. If you want to change the message, cancel it and request a revision in one sentence (“make it shorter, do not mention Friday”), and the Assistant will return a new suggestion. The email is sent only after you click the button. It is sent from your personal mailbox (SMTP under Settings → Profile), so the client sees your address as the sender, and their reply goes to your inbox, not the application. Step by step: Send email.
Before clicking, check the meeting date, the address and whether “2 days” comes from the agreement rather than the model’s assumption. Three seconds that can save you embarrassment.
A more complex thread? Turn on Expert mode
Sometimes the problem is not writing but reading: the Nord thread has 40 messages, someone changed their mind halfway through, and you do not know what is current.
The “Organise” button above the thread has four reading modes. The fourth, Expert, rewrites messages using professional language and adds exactly one “Expert note” line to each one - advice, a risk or a question worth asking. This is the only place where the model can add something of its own, which is why it is labelled separately. For the Nord proposal, a note might say: “The client is asking for 30-day payment terms, but 14 days were agreed in June - confirm before sending.” The result only lasts for the session, and the original messages remain unchanged.
| Mode | What it does | When to use it |
|---|---|---|
| Original content | turns off AI | when you want to see exactly what was written |
| Polish | the same content, shortened and written in the first person | long messages written in a hurry |
| Summarise | 1-2 sentences per message | a thread with 40 messages when you need an overview of what happened |
| Expert | professional language + one “Expert note” | a proposal, contract or complaint - wherever mistakes are costly |
The screenshot below shows Expert mode enabled in the “Website project” thread: above the conversation is a banner saying “You are reading the Expert version prepared by AI”, and under each message from Ola Nowak and Magda Wiśniewska is a single “Expert note” line - about the mobile version of the mock-ups, the criteria for choosing between two options and the logo in SVG format.
AI Agent: when the steps need to happen automatically
The plan from the first scenario has one drawback: you have to remember to follow up for a response. That is what the AI Agent is for. Back to the Nord thread: Kasia wrote, “We are waiting for a decision by Friday”, and Ola writes in the same thread, in private mode:
“If Kasia has not replied by tomorrow, remind her at 8:30 and let me know at 9:00.”
The Agent turns this into a plan - you can see it in the opening image for this article: a “AI Plan #16 saved” card with a summary (“I will wait for Kasia’s reply until 8:30 tomorrow. If there is no reply, I will remind her in the thread and let you know at 9:00”) and three steps with dates and times: wait for a reply, send a message at 08:30, notify at 09:00. Below the plan is the command for cancelling the entire plan (“@aipriv /anuluj plan 16”). In the AI Assistant window, before anything is saved, you see a card with the plan and a “Yes, proceed” button. Once approved, the steps appear in the Calendar as graphite-coloured entries with a robot icon next to the tasks. Clicking one opens a window where you can change the time or cancel an individual step - the rest of the plan will continue as scheduled.
What you need to know before handing work over to the AI Agent:
- A plan can have up to 10 steps and expires after 30 days.
- Without a CRON job on the server, plan steps are carried out whenever anyone on the team has the application open; with CRON configured once by an administrator, they run on time, including at night and at weekends. The application displays a warning when you schedule a step if CRON is not configured.
- The AI Agent writes on your behalf and within the limits of your permissions.
- If the AI Assistant says “I set a reminder” but no reminder was created, you will see a yellow “Correction” box below the response. The application makes sure AI does not take credit for work it has not done.
Where AI ends and your responsibility begins
The AI Assistant suggests, but does not decide. A plan for Friday, a 10% discount or a payment due date - these are suggestions based on what it found in the sources. If Marta wrote about a different client in June, the answer will sound just as confident. Check all figures, dates and names.
The AI Assistant can only see text. It can read PDF, DOCX, XLSX, CSV and text files, but a scanned contract without a text layer will appear blank to it - the application does not have OCR.
The AI Assistant does not know what you have not shown it. No model can recreate arrangements made over the phone if nobody wrote them down. One sentence in the thread after the call is enough (“Piotr confirmed the implementation, 2 days, starting in October”).
The application also works without AI. Tasks, to-do lists, Calendar and Notes do not require an API key. AI is enabled by an administrator - separately for tasks, chats and the AI Assistant window.
How to get started with your team
- The administrator connects the key. Settings → AI: provider, API key and a model from the list retrieved from the account.
- Everyone tries it on one of their own tasks. Ideally, choose one that has not moved forward for two days: click “AI Assistant” in the thread and ask “where should I start?”
- Record decisions where AI can find them. A note with your price list, a folder containing proposals, or a sentence in the thread after a phone call. The more knowledge you keep in the application, the fewer generic answers you will get.
The task “Proposal for client Nord” still needs you to write it. The difference is that at 9:40 on Tuesday, you do not start by searching for information, but with a plan based on your team's own arrangements. If you want to try this with sample projects, leave your contact details in the Demo - we will send you access to the demo version.
Frequently asked questions
Can the AI Assistant see other people's tasks?
It responds within the permissions of the person asking: a regular user only gets information they can see in the application, while an administrator has access to all tasks and chats. No one - including an administrator - can access someone else's private notes, files from private folders or mail.
Can other task participants see a question asked to the Assistant in a thread?
No. “AI Assistant” mode in a thread is private: the question and answer remain in the task with a “FOR YOU ONLY” label. A task created using “Save” or an email sent after confirmation is a normal, visible action.
Can AI send an email or create a task without my knowledge?
No. An email is sent after you click “Send”, a task is created after you click “Save” in the open form, and an Agent plan starts after you select “Yes, proceed”. If a response claims that something has been done when it has not, the application adds a “Correction” section.
Where does my data go when I ask the Assistant a question?
Application data is stored on your server. The request, together with the context needed to answer it (parts of threads, notes and files), is sent to the model provider selected by your company (OpenAI, Google Gemini, Anthropic Claude or xAI Grok) using its own API key, with no intermediary mark-up. Mail as a knowledge source is disabled by default - an administrator can enable it.
Sources
- Microsoft, Work Trend Index 2024: “AI at Work Is Here. Now Comes the Hard Part” (survey of 31 000 employees in 31 countries, February - March 2024)
- Task Manager user guide: AI Assistant (private mode in a thread, knowledge sources, checklists, sending emails, AI Agent, Correction)
- Task Manager user guide: Calendar (AI Agent messages on the grid, cancelling an individual step)
The Task Manager Team - we write about teamwork, AI in business and keeping tasks organised, using examples from Task Manager, software installed on your own server. Task Manager has been developed by Blue Creation Group (Piotr Łącz) since 2013 and is used by organisations including Neonet, the Military University of Technology and Polpharma - read customer reviews. We check our articles against the user guide and the latest version of the software.
