On Wednesday, Ola returns from a trip and opens the "Client arrangements" chat: 14 messages from three days are waiting there. Instead of reconstructing the conversation from the start, she asks the AI Assistant about decisions, deadlines and who's responsible. AI at work starts with a specific matter like this.
AI at work makes sense where the matter actually lies
The easiest place to start is with tasks where you already have the text at hand: a conversation, a task description or an attachment. In Task Manager, you ask the AI Assistant within the same thread, instead of copying the arrangements into a separate chat.
The instruction "use AI" alone changes little if you have to reconstruct the context before every question. Within a task or chat, the Assistant sees the content of the thread you have access to and its attachments. When working in the separate AI Assistant window, you can use the "Sources" button to point to a task, chat, note, Workspace or calendar. You choose which matter you're asking about; the Assistant doesn't gain access to other people's tasks just because it knows their names.
Try a simple test: list five tasks from the past week that mainly involved reading, rewriting or keeping track. Don't write "choosing an offer for the client". Write "gathering three arrangements from a conversation before choosing an offer". For a broader perspective aimed at business owners, see AI in business without the hype.
How to read a long thread and an attachment? Four tasks
You can ask the AI Assistant for a summary of a conversation, or use the "Summarise" mode to shorten each message individually. The "Expert" mode will highlight points that need clarifying, and asking about an attachment will help you find a specific passage in a document.
1. Summarise the conversation. On Wednesday, Ola clicks "AI Assistant" under the chat and types: "Summarise this thread". She gets the arrangements split into decisions, deadlines and people responsible. Only she sees the question and the answer - the answer is marked with a "FOR YOU ONLY" badge. She can still check the original messages if an arrangement needs verifying.
2. Shorten messages without replacing them with a single summary. On Thursday, Marek selects "Organise" and then "Summarise". Each message gets a version in one or two sentences. The Assistant processes the eight most recent messages first; you can add older ones using the "Organise older messages (N)" button. This is a preview for the duration of the session, not a change to the saved conversation. Below the processed text there's a warning: "AI can make mistakes".
3. Spot a question you might have missed. On Friday, Marek is preparing for a discussion about mock-ups and selects the "Expert" mode. An "Expert note" appears next to messages - advice, a risk, or a point to clarify. It's not a decision made for Marek. It's a reason to check, before the call, whether the client has defined the criteria for choosing the mock-up.
4. Ask about an attachment. When a contract in DOCX is attached to a task, you can ask for the passages relating to the deadline to be pointed out. The Assistant also reads, among others, TXT, CSV, XLSX and PDF with text; it retrieves long documents in parts. It won't read an old binary DOC or a PDF scan without a text layer, as it has no OCR, i.e. recognising text from an image. For such a scan, ask for a file with text, or paste the content into your question.
How to write with AI and still check the text? Four tasks
You can improve your own message, prepare an email draft, translate a reply, or compile content into a downloadable file. You check every text before sending it; in Task Manager, an edit in the editor doesn't send the message, and an email prepared by the AI Assistant requires your confirmation.
5. Improve your own message. On Tuesday, Kasia writes three paragraphs to the team about a changed deadline and clicks the "Polish text" wand icon in the editor. The feature improves the language while keeping paragraphs, lists and bold text; it doesn't add new facts. Kasia reads the result, and if she prefers the original wording, she clicks "Text polished - click to undo". Nothing reaches the conversation until she sends the message herself.
6. Prepare an email to a client. On Thursday, Kasia asks for a message giving the deadline for sending the mock-ups. The Assistant shows a draft, which she can read and confirm; it doesn't send the email itself. Sending uses your connected inbox in the Mail module, available only in the Pro plan. Without a connected inbox, the Assistant will report that it's missing and won't send the message.
7. Translate a reply within its own thread. On Friday, Ola reads an English message about the "Offer for client Nord" and asks the AI Assistant for a Polish version. The translation appears as a private reply next to that conversation. If the wording of the offer hinges on a single word, Ola still checks it against the original.
8. Turn the content into a downloadable file. Instead of copying the task summary into a document, just ask: "Prepare a Word document with a summary of this thread." The Assistant writes the content, and Task Manager puts together a DOCX file and shows a "Download …" button. PDF, XLSX and CSV are also available; spreadsheets require data arranged in a table. Before sending the file, check the names, numbers and dates.
How to organise descriptions and data? Three tasks
AI can turn a description into a checklist, data into a table, and a sentence into a completed task form. You check the list items and the numbers in the table, and correct the new task's fields before clicking "Save".
9. Turn a description into a checklist. On Monday, Magda opens the task "Website project" and asks: "Turn the task description into a checklist I can tick off." The Assistant returns five items; after completing two, the bar shows "2 of 5 done". A list from a private conversation with the Assistant is visible only to Magda. When a clickable list is in a regular message available to the team, the ticks are shared and you can check who ticked off which point. You can read more about what should be a list item versus a separate task in the article about the checklist in a task.
10. Build a table from data. On Wednesday, the owner asks: "Make a table comparing Q1 and Q2 costs." The Assistant can show it in the reply; if a spreadsheet is needed, you then ask for Excel or CSV. First, though, check whether you've pointed to the right sources. Beneath the list of sources you'll see a usage bar, and if the model can't take in all of the selected context, the Assistant will tell you what it left out.
11. Prepare a new task from a single sentence. On Friday, Magda asks in chat: "Create a task: send photos for the Nord offer, due Friday, assign to Marek." The Assistant opens an "Add task" window, visible only to her, with the fields already filled in, including the Workspace of the current thread and the due date converted to an actual date. Magda can edit the form. The task is only created once she clicks "Save"; closing the window creates nothing.
Keeping watch: a reminder that follows the due date
A reminder can be linked to a task's due date rather than just a date entered once. When you change a task's due date in Task Manager, the reminder pinned to it moves along with it.
12. Pin a reminder to a task. On Tuesday, Ola writes: "Remind me about this task a day before the due date." It's only saved once she clicks "Yes, do it" in the confirmation window. At the set time, she'll get a notification in the app; if the task's due date has no time set, the program assumes 9:00. Tasks without a due date can't be watched this way - the Assistant will suggest a regular reminder instead. The limit is 50 active reminders.
If you want to receive an email while the app is closed, CRON must be running on the server - a recurring job triggered by the hosting. If CRON has never started, the app warns you about it when you set a reminder. Separately, you can enable a daily summary of your own tasks due today and overdue in your Profile; the default time is 8:00.
Besides the list of 12 tasks, there's also private activity history. There's also a question that usually comes back on a Friday: what did I actually get done over the last few days? If you turn on private activity history, the program records the titles and links to what you've created, changed and viewed; you can see recent events in Settings → Profile → Activity History ("Show recent events"). The log doesn't store message content or notes, and entries older than 90 days are deleted. No one but you, not even the administrator, can see this history; however, a summary of it is sent to the configured AI provider when you use the Assistant. You'll find details in the article about activity history and the work-day report.
Which decisions shouldn't you hand over to AI?
Don't hand over decisions about priorities, or responsibility for the accuracy of information sent to a client, to AI. The Assistant helps you read through material and prepare an action, but with a very broad question it may only pull in a handful of the best-matching items at once.
When the answer is meant to concern a specific matter, point to it under "Sources" rather than asking about "everything". Also check the result before using it. If the Assistant confirms an action it hasn't actually carried out, Task Manager will show a yellow "Correction" box under the reply describing the actual state of things.
The privacy of a thread doesn't mean the query stays solely on your server. The content of the question, together with the context used, is sent to the chosen model provider. Before turning on AI at your company, decide which matters the team will entrust to it; this is covered in more detail in the article what gets sent to the AI provider.
Where should a small business start?
Start with a single task whose result is easy to compare with the original: summarising a conversation, or improving your own message. Task Manager also works without AI, so the administrator can enable it separately for tasks, chats and the Assistant window, instead of changing the whole company's way of working at once.
The administrator can choose between OpenAI, Google Gemini, Anthropic Claude and xAI Grok. The company provides its own API key and pays the model provider directly, without a middleman or a Task Manager markup. The article about your own AI key at the company explains how to approach this configuration.
The application's data sits on the client's own server, and the licence is perpetual: you pay once for a number of seats, with no per-person subscription. It includes 12 months of updates, after which renewal is optional. For the Classic plan for up to 20 people, the price is PLN 1,490 for the Standard variant or PLN 3,990 for Pro. Mail and the Files module belong to Pro; the AI Assistant is available in both variants.
Tomorrow, open the longest thread you have to read and ask: "Summarise this thread." Compare the answer with the messages before you pass the conclusions on to anyone. If you're looking for a place for your team where AI works right alongside tasks, where the application's data sits on your own server, and the licence doesn't charge a fee per person every month, take a look at the Demo.
Frequently asked questions
Do you have to pay separately for AI at work?
Yes - when using AI in Task Manager, you pay your chosen model provider for using your own API key. Task Manager itself also works without AI and doesn't add a mark-up to requests.
Can AI send an email or create a task on its own?
No - an email requires your confirmation, and a task is only created once you click "Save" in the form. A reminder also requires confirmation via the "Yes, do it" button.
Will others see what I ask the AI Assistant within a task?
No - the question and answer in private mode are visible only to the author. Conversations in the AI Assistant window are private even from the administrator.
Will the AI Assistant read a scanned PDF contract?
No - a PDF scan without a text layer won't be read, as the program has no OCR. A PDF containing text and a DOCX document can be used as a source of answers.
Does the "Summarise" mode change chat messages?
No - the "Summarise" mode shows a processed version only for the duration of the session. The original messages remain untouched.
Sources
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.
