How to build a “Meeting Brain” AI agent in a weekend (no code)

I’ve been experimenting with AI agents for real productivity use cases lately. Here’s one you can build in a weekend — no engineering degree required.

It automatically captures Zoom transcripts, pulls out action items and key decisions, and sends follow ups via email or Slack.

Here’s exactly how to build it.

Step 1: Automatically record and transcribe your calls

Tool: Grain (or Fathom)

Setup:

• Sign up at Grain, connect your Zoom account.

• In settings, turn on “Auto-Join & Transcribe” for all meetings.

• Go to Integrations → create an API key. Save it.

• Now every Zoom call is recorded, transcribed, and available via the Grain API within minutes.

Optional: You can poll their API or use their webhook to get notified when a new meeting + transcript is ready.

Step 2:  Set up a Make.com scenario to fetch the transcript

Tool: Make (or Zapier — but Make gives you more control)

Setup:

• Create a Make account and a new scenario.

• Add an HTTP GET module: https://api.grain.com/v1/recordings → Set auth header: Authorization: Bearer YOUR_API_KEY

• This returns recent Zoom recordings — each with a title, timestamp, and recording ID.

• Filter to only include new recordings (e.g. last 10 minutes) to avoid duplicates.

• Extract the id field from each new recording.

• Add another HTTP GET module: https://api.grain.com/v1/recordings/{id}/transcript

This returns the full transcript text — ready for GPT.

Step 3: Send transcript to GPT (agent logic)

Tool: GPT-4 or Claude

• Add another HTTP module to Make — this time for OpenAI’s chat endpoint.

• Use gpt-4 (or gpt-3.5-turbo for cost/speed).

Prompt setup:

• System Prompt: You are a meeting assistant AI that extracts clear action items and decisions from meeting transcripts.

• User Prompt: Here is a transcript from a Zoom meeting. Extract only actionable items, clearly assigning each task to the right person if possible. Include due dates if mentioned. Then summarize any key decisions made. Only output plain text. No filler, no greetings.

• Pass the transcript from Step 2 as the user message content.

GPT will return something like:

• John to send the revised pitch deck by Friday  

• Sarah will draft onboarding emails by Tuesday  

• Decision: Move launch date to May 10

Store that result in a variable for the next step.

Step 4: Auto-send follow up (Email or Slack)

Now let’s actually take the action — this is what makes it a true agent. 

Choose how you want your agent to follow up — email, Slack, or both.

Gmail:

1. Add Gmail to your Make scenario.

2. Use the “Send email” module.

3. To field: Use the meeting attendee emails (you can get these from Grain’s metadata).

4. Subject: Follow up from [Meeting Title]

5. Body: Paste in the GPT output directly.

6. Done.

Slack:

1. Add a Slack module.

2. Choose the channel or user(s) to notify.

3. Send the same text GPT generated in Step 3.

Optional: Add memory (context across meetings)

If you want your agent to feel smarter over time — add a simple memory layer.

How:

1. Create an Airtable with columns: Date, Meeting Title, Participants, Action Items, Decisions, Completed?

2. In Make, after sending the message (email or Slack), add a new step:

→ Use the Airtable “Create Record” module to insert a row with data from this meeting.

Now you’ve got memory — a running log of what happened in every meeting.

Bonus:

1. Automatically send reminders if action items are still open after X days.

2. Feed past meeting data into GPT with a prompt like: “Here’s what we discussed last week. Update the plan based on today’s call.”

That’s it. You now have a meeting AI agent that listens, thinks, and acts.