According to Statistics Poland (GUS), in 2025, artificial intelligence technologies were used by 8,7% of businesses employing at least 10 people, and by 6,1% of small businesses (10-49 employees). The rest mainly know AI from presentations and headlines. This article shows seven things the AI Assistant in Task Manager can do during a team’s normal working day: what you tell it, what you get and how much it costs.
How AI built into your work differs from a separate chatbot
Most companies that have “tried AI” opened a chat in their browser, pasted in an email and then pasted the response back. The model does not know who is who on the team, what the due date is or what was agreed a week ago. In Task Manager, the AI Assistant works where the work happens: you can ask about an open thread or select a Workspace, note or file folder as a source. You choose the model yourself from OpenAI, Google Gemini, Anthropic Claude and xAI Grok - the difference is not the model, but the fact that the AI can see the context and take action in the application.
Below are seven situations involving three teams: Bilans Plus, an accountancy firm in Poznań (7 people), Studio Kreska, an agency in Wrocław (12 people), and Termo-Serwis, a service company near Łódź (9 people). The companies are fictional, but the instructions and results are real.
1. Summarising a long thread
Situation. Ola from Bilans Plus returns from leave. The task “Quote for client Nord” has 46 messages waiting for her: the client changed the scope, Marek asked about VAT and someone uploaded three versions of the contract.
Instruction. In the thread, she clicks Organise and selects Summarise. Alternatively, she switches the message field to question mode (the AI Assistant button) and writes: “Summarise this thread - what was agreed, who is responsible for what and what due dates were mentioned”.
What she gets. In Summarise mode, each message is turned into 1-2 sentences, processed in batches of 8 starting with the latest. In question mode, she gets a single summary divided into decisions, due dates and “who is responsible for what”, visible only to her. The original messages remain untouched, and every processed message includes the note “AI can make mistakes”. It takes a few seconds instead of half an hour of reading.
The screenshot below shows a shorter example from the demo: the “Client arrangements” chat, with 14 messages from three days. The Assistant extracted three decisions about the website (a dark colour palette, a contact form and a blog in the second phase), two due dates (feedback on the mock-up tomorrow by 15:00 and a meeting on Friday, 25 September at 11:00) and the division of work between Marek, Anna and Ola - along with two quote options, 14 900 and 19 800 zł.
More: Summaries.
2. A checklist from a task description
Situation. At Studio Kreska, Michał creates a task called “Website project” for client ABC. The description is two pages long: page structure, copy in two languages, mock-ups, coding, uploading the content and final approval.
Instruction. In the task thread, he enables question mode (the AI Assistant button) and writes: “Turn the task description into a checklist I can tick off”.
What he gets. A list of items with checkboxes and a progress bar, visible only to him - conversations with the Assistant in a thread are private. The screenshot below shows five items, two of which have already been ticked and crossed out (the structure of five pages agreed with client ABC, PL and EN copy collected), while the bar shows “2 of 5 completed”. To let the entire team work with the list, Michał copies the items into the task description field and marks them using the To-do list button (the same one used in the message editor). From then on, the checklist is interactive on the task card and its status is shared: when Ania ticks off the mock-ups, everyone else sees it immediately, along with her mini avatar and the date. Fifteen minutes spent breaking down a brief is replaced by a single sentence and a quick paste.
More: Checklists.
3. Turning an email into a task
Situation. Termo-Serwis receives complaints and problem reports by email, at different times and in different formats. Until now, someone would check the inbox each morning and copy them into a list.
Prompt. The owner gives the AI Assistant a rule: “When a new email arrives in the inbox and AI determines that it is a complaint or problem report, create a task in the Sales workspace, assign it to Kasia Wójcik, set the priority to high and the due date to tomorrow”. The Assistant creates a draft WHEN - IF - DO Recipe and displays a “Go to Automations” card. The owner corrects one field, clicks Save, then Enable.
What they get. Every email that AI identifies as a complaint becomes a task with an assignee and due date before anyone opens the inbox. The “AI determines” condition can see the sender, subject, a content preview and attachment names; if uncertain, it answers “no” - the automation would rather do nothing than do the wrong thing. To reduce the cost of queries, you can add cheaper conditions before it (such as a specific inbox or a word in the subject) - AI only evaluates messages that pass those conditions. The Test button shows what would happen without taking any action. This scenario requires Mail and Automations (the Pro version).
The “Customer complaint” Recipe from the demo, still disabled, just after being saved: the AI condition written in plain language (“The message is a complaint or problem report - the sender is dissatisfied, something is not working, or they are requesting a repair, refund or explanation”) and the “Create task” action with chips for Sales, assignee: Kasia Wójcik, priority: high, due date: +1 days.
The full scenario, step by step: An email that creates a task automatically.
4. A table of tasks and export to Excel
Situation. The management team at Studio Kreska wants a monthly breakdown of completed tasks by client. Michał used to prepare it in a spreadsheet by copying items from the list.
Prompt. In the Assistant panel, with the Clients workspace selected: “Create a table of tasks completed in September: client, subject, who completed it, due date, and whether it was overdue”. After checking it: “Generate an Excel file from this”.
What they get. A table directly in the response, followed by a “Download” button with an .xlsx file after the second prompt. Word, PDF, CSV and HTML work in the same way - AI writes the content and the application creates the file. The data comes exclusively from tasks that the person asking the question can access; for very broad queries, narrow down the date range or workspace.
More: Tables and reports.
5. Translating a message in a Thread
Situation. Termo-Serwis maintains a facility for a German tenant. The facility manager writes in German, while the technicians reply in Polish.
Prompt. In the Thread, in question mode: “Translate the latest message into Polish”. Once the reply has been agreed: “Translate into German: valve replacement on Friday at 8:00, access to the plant room required”.
What they get. The translation appears next to the original and is visible only to the person asking the question (with an “ONLY FOR YOU” label). They copy the version they want to send into a regular message and decide what is sent to the client. The model knows the context of the Thread, so “valve” is not mistranslated as “air valve”.
6. A Reminder from a single sentence
Situation. Ola must send the client’s declaration by the 25th; the Task is due on the 24th. She wants a Reminder but does not want to click through the Calendar.
Prompt. In the Task Thread: “Remind me about this Task one day before the due date”. Or in the Assistant panel: “Remind me tomorrow at 9 about the invoice for Nord and send it by email as well”.
What they get. In the Task Thread, a Reminder linked to the due date is created immediately - the application calculates the time, and the Assistant replies “Reminder set: one day before the due date” with the date and time. If the due date changes to the 28th, the Reminder automatically moves to the 27th. In the panel, when you specify a particular date, you first get a card with the date and click “Yes, do it” - only then is the Reminder created. An in-app notification arrives at the specified time, while an email is sent only if explicitly requested. You can check the list by saying “show my reminders” and cancel one you no longer need by saying “cancel reminder” followed by its number from the list.
7. AI Agent: a plan instead of a single answer
Situation. In the “Quote for client Nord” Thread, Ola from Bilans Plus is waiting for a reply from Kasia Wójcik. She cannot finalise the quote without it and does not want to keep checking the Thread.
Prompt. In the Task Thread, in question mode: “If Kasia has not replied by tomorrow, remind her at 8:30 and let me know at 9:00”.
What they get. The AI Agent saves the plan immediately and displays it in the Thread as an entry titled “AI Plan #16 saved”, with a one-sentence summary (“I’ll wait for Kasia’s reply until tomorrow at 8:30. If she has not replied, I’ll remind her in the Thread and let you know at 9:00.”) and three dated steps: (1) wait for a reply until 8:30, (2) send a message at 8:30, (3) send a notification at 9:00. Steps scheduled for now run immediately, while those scheduled for later are triggered by a cron job on the server. If Kasia replies at 16:00, the conditional steps are skipped and the plan ends automatically. A ready-to-use cancellation command appears below the entry: “@aipriv /cancel plan 16”, which can be used at any time. The Agent operates within Ola’s permissions and only among the Thread participants.
How the AI Agent differs from the Assistant: AI Agent.
What requires confirmation and what does not
| Action | Confirmation |
|---|---|
| Summary, table, translation, Checklist | no - the response itself does not change anything |
| Task from the Assistant | yes - Save in the Task window |
| Email from the Assistant | yes - “Send” |
| Automation rule | yes - Save and Enable |
| Reminder from the panel | yes - “Yes, do it” |
| Reminder from a Thread (linked to the due date) | no - it is created immediately; cancel it by saying “cancel reminder” |
| AI Agent plan | no - the plan is saved immediately; cancel it with a single sentence |
If the model says “I’ve set a reminder” but has not done so, a yellow “Correction” block will appear below the response, explaining what actually happened.
How much it costs: your own key, no mark-up
Task Manager does not sell AI packages. You buy the application licence once: the Standard version costs Starter 690 zł excl. VAT (up to 8 people), Classic 1 490 zł (9-20 people) and Max 2 290 zł (21-35 people); the Pro version with Mail, Files and Automations, required for scenario 3, costs 1 990 / 3 990 / 5 990 zł excl. VAT (see the full pricing). You pay the model provider for queries according to its pricing, using your own API key, with no mark-up and no “queries included in the package” limit.
Providers charge for tokens, which are chunks of text (according to Anthropic, around 4 characters in English; Polish uses more). Selected rates per million tokens from official price lists (USD, as of 20 September 2026):
| Provider and model | Input (1 million tokens) | Output (1 million tokens) |
|---|---|---|
| OpenAI gpt-5-mini | 0,25 USD | 2,00 USD |
| OpenAI gpt-4.1 | 2,00 USD | 8,00 USD |
| Anthropic Claude Haiku 4.5 | 1,00 USD | 5,00 USD |
| Anthropic Claude Sonnet 5 | 2,00 USD | 10,00 USD |
| Google Gemini 3.5 Flash-Lite | 0,30 USD | 2,50 USD |
| Google Gemini 3.5 Flash | 1,50 USD | 9,00 USD |
The example from section 1: a thread with 46 messages might use 8 000 input tokens, while the summary uses 600 output tokens. With gpt-5-mini, that costs around 0,003 USD (8 000 × 0,25 / 1 million + 600 × 2 / 1 million), with Claude Haiku 4.5 around 0,011 USD, and with gpt-4.1, Claude Sonnet 5 or Gemini 3.5 Flash around 0,02 USD - two cents. A team of 10 people asking 10 questions a day over 22 working days generates 2 200 requests a month: around 44 USD at two cents per request, or a few dollars with a “mini” model. This is an estimate: the bill depends on the model, thread length and number of sources; you set the spending limit in the provider's dashboard. We have not included xAI's pricing for Grok because we did not verify it on the publication date.
Billing details: AI costs with no mark-up.
Where your data goes
Application data - tasks, messages, files and mail - stays on your server. But when you ask for a thread summary, its content (and any selected attachments) is sent in a request to your chosen model provider using your key. What the software does to keep you in control:
- The administrator decides what can be used as a source. Mail is disabled as an AI knowledge source by default because correspondence contains third-party data; once enabled, each person can only see their own mailboxes.
- The model only sees what you have access to. The Assistant cannot access a Workspace you do not belong to; documents in the knowledge base are filtered according to permissions. The exception is the administrator, who can use the Assistant to see all tasks and chats (but not other people's mail).
- The API key is encrypted and is no longer displayed in full once saved.
- The source usage bar shows how much content fitted into the request.
Before enabling AI for your team, read the chosen provider's data processing terms - you enter into an agreement with them through your API key. You can change the provider at any time under Settings → AI.
The seven activities described here require no implementation or change to how you work - just an API key and one sentence in the right place. The easiest way to try them is with your own threads: Demo.
Frequently asked questions
Can AI do anything in the software without my knowledge?
Not without your knowledge - but in two cases, it can happen without a click. A task, email, reminder created from the panel and an automation rule wait for your confirmation. An AI Agent plan and a reminder linked to a due date are saved as soon as you enter your instruction, but they leave an entry in the thread and you can cancel them with a single command (“@aipriv /cancel plan 16”, “cancel reminder”).
Which model should you choose to start with?
A cheaper model from the “mini” or “Flash” family is enough for summaries, checklists and translations. A more expensive model makes sense for long documents and reports where precision matters. You can change the model with one click in Settings → AI.
Is data from my server sent to the AI provider?
Yes - to the extent that you select it as a source for the conversation or it is covered by an automation rule. Your application data remains on your server; a request containing the context is sent to the provider using your key and subject to that provider’s terms.
Will the AI Agent continue working when I close my browser?
Running plan steps and sending emails in the background requires a scheduled task (cron) on the server; hosting providers that support HTTP cron are also suitable. Without it, plans only start when someone on the team is using the application.
Sources
- Statistics Poland, “Information society in Poland in 2025” (enterprises with 10 or more employees; pp. 112-114)
- OpenAI, “Pricing” (API)
- Anthropic, “Pricing” (Claude Developer Platform; also the estimate of 1 token ≈ 4 characters)
- Google, “Gemini Developer API pricing”
- Task Manager, pricing
- Task Manager user guide: AI Assistant, AI Settings, Automations - https://taskmanager.pl/materialy/asystent-ai, https://taskmanager.pl/materialy/ustawienia-ai, https://taskmanager.pl/materialy/automatyzacje (accessed: 2026-09-20)
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.
