Why reports take up so much of a manager's time
Information gets collected by hand. All week, people post in threads about what they finished and what's blocking them. On Friday the manager asks everyone to send a recap, and they retell the same things a second time. One person sends a spreadsheet, another writes a DM, and all of it has to be turned into one text.
Meeting outcomes don't get written down. The stand-up agreed on deadlines, and a week later people remember the decision differently. Often nobody takes notes.
Hundreds of unread messages after a vacation. To figure out what changed, the manager scrolls through chats for half a day and still misses a question that was waiting for their answer.
An AI chat can't see your messages. For ChatGPT or another assistant to write a report, you have to copy the messages into it. That takes time, and the assistant only sees what you pasted: a thread you forgot about won't make it into the report.
What to set up in Pachca
Call transcripts and AI summaries. Pachca transcribes calls in chats and channels on its own. After a call, the transcript and an AI summary arrive in the thread under the call message, with nothing to set up. To set your own prompt, click the arrow next to the video call button in the chat header and choose Configure AI summary. The prompt applies to every call in that chat. You can see how it looks in the use case "How do you record a meeting and get notes automatically?".
A personal AI agent. Claude Code, Cursor, Codex or another agent connects to Pachca through the Pachca CLI. Install the CLI, log in with a token from Automations → API and add Pachca skills to the agent with a single command. The agent sees only the chats you see. You can revoke access at any time with pachca auth logout or by deleting the token. Step-by-step instructions are in the article "How to connect your personal AI agent to Pachca".
Drafts and scheduled messages. The agent can create a draft in any chat or thread and schedule a message for a set time, such as a Monday reminder for the team. This only works with a personal token, and a chat can hold up to 50 scheduled messages.
An agent for the whole team. If several managers need reports, you can set up a team agent, which is a bot: people mention it in a thread and it replies there under its own name. We showed how to build one in the use case "How do you build an AI agent without code?".
If people rarely post about work in the chat. The agent has nothing to build a report from if the team doesn't post about their tasks in the chat. In that case a bot collects statuses with a form before the stand-up, as in the use case "How do you run a daily standup in 15 minutes?".
Five tasks a manager can hand to AI
An agent connected to Pachca reads the same chats and threads you do. So the prompt only needs to name the chat and the time period. You don't have to retell the conversation.
A weekly report for leadership. The agent goes through the department chat's threads for the week and pulls together what got done, what was pushed back and where help is needed. You check the numbers and send it.
Read the [Team] Sales chat and its threads for September 22–26. Write a report for the director: what we did, what we didn't finish and why, and where we need their decision. One line per item, with a link to the thread for each. Don't round or guess numbers: if they aren't in the messages, say so.
Meeting recaps. After a call in a chat or channel, Pachca posts an AI summary and a transcript to the thread under the call message. The summary prompt decides what goes into the recap. You set it once per chat, and it then applies to every call there.
Summarize the call: 2–3 sentences with the key points, then "Decisions" and "Tasks" with owners and deadlines. If no deadline was named, write "no deadline named".
A plan for the week. The agent turns the stand-up summary and the threads into tasks with owners and deadlines and prepares a message for the chat. If nobody named a deadline or an owner in the meeting, the agent marks it as an open question.
Take yesterday's stand-up summary from the call thread in [Team] Sales. Write a plan for the week: task, owner, deadline. Put it in this chat as a draft, I'll send it myself. Schedule a message for Monday at 9:00 reminding the team about the tasks with no deadline.
Thread
[Team] Sales
Catching up after a vacation. The agent retells what happened in your chats over the dates you give it and lists separately where people are waiting for your reply.
Look through my chats from September 8 to 19. Tell me what changed in our projects, what decisions were made without me and where people are waiting for my reply. Put the questions for me first.
Preparing for a one-on-one. Before a conversation with a team member, the agent gathers what they worked on over the month, where they asked for help and what you agreed on in your messages.
Find what Daniel Reed posted in the department chats in September. List his tasks and their status, where he asked for help and what we agreed on in direct messages. No judgments, just facts with links.
How to check what the AI wrote
Check the numbers. Speech recognition makes mistakes with names, terms and numbers. Check the amounts, deadlines and names in the report against the transcript or the CRM.
Ask for links to threads. A report item with a link takes a minute to verify: open the thread and read what was actually said.
Send it yourself. Ask the agent to put the text in the right chat as a draft. The draft shows up in the message field and in the Drafts section. If the agent sends a message itself, it goes out under your name. A chat holds one draft, so the agent will replace any unsent text of yours.
Let it say "no data". Without that instruction in the prompt, the agent may fill a gap with something plausible it made up. Add a line to the prompt: "Don't guess. If there's no data, say so."
Where to start
Start with one report you write every week. Connect a personal agent and ask it to build that report from the department chat using a prompt from this article. Compare it with what you would have written yourself, and add whatever the agent missed to the prompt.
Frequently asked questions
The agent takes numbers from the messages and can get them wrong if people wrote them differently in threads or didn't write them at all. Ask it to link a thread for every number, and check amounts against the CRM or your accounting system before sending.
No. A personal agent works with your token and sees only the chats and threads you can see. A bot agent sees public channels and, of the private chats, only those it was added to as a member.


