Most meeting agendas die in a Slack message that says “what are we covering tomorrow?” Nobody answers, the meeting wanders, and someone spends the first ten minutes figuring out why everyone is in the room. A small AI agent fixes this quietly: it watches your calendar, pulls in the relevant context, drafts a sharp agenda, and drops it where the team already looks. No code, no developer, about an hour to set up.
This guide walks through building that agent with no-code tools. We build these for real, so I’ll be specific about what to wire up, what to put in the prompt, and where the cheap version falls apart.
What this agent actually does
Let’s be precise, because “AI agent” gets used loosely. You’re building an automation that does four things in sequence:
- Triggers on an upcoming calendar event (say, 24 hours before, or when an invite lands).
- Gathers context — the invite title, attendees, the description, and ideally the last meeting’s notes or an open thread.
- Drafts the agenda by sending that context to an LLM with a tight prompt.
- Delivers it — pastes it into the calendar event, posts to a Slack channel, or emails the attendees.
Whether you call it an “agent” or a “workflow” matters less than getting steps 2 and 4 right. The LLM in the middle is the easy part. The value is in feeding it good context and putting the output somewhere people see it.
Pick your platform
You have three realistic no-code routes. They differ mostly in how much glue work they hide from you and what they cost.
| Tool | Best for | Watch out for | Rough cost |
|---|---|---|---|
| Zapier | Fastest setup; huge app library; “AI by Zapier” step needs no API key | Multi-step Zaps with AI eat tasks quickly; complex logic gets awkward | ~$20–50/mo at usable volume |
| Make | Visual flows, branching, loops; cheaper per operation | Steeper learning curve; you’ll manage your own OpenAI/Anthropic key | ~$9–30/mo |
| n8n | Most control, self-hostable, no per-task ceiling | Closest to “technical”; you maintain it | Free self-hosted / ~$20+ cloud |
Honest recommendation: if this is your first agent, start in Zapier. The built-in AI step means you don’t touch an API key, and the Google Calendar/Outlook and Slack connectors are mature. Move to Make or n8n when you want branching logic (different agenda templates per meeting type) or when Zapier’s task pricing starts to sting. Don’t reach for n8n on day one unless you’re comfortable babysitting a self-hosted service.
Build it step by step (Zapier example)
1. Set the trigger
Use Google Calendar’s “Event Start” trigger with a lead time of 24 hours, or “New Event Matching Search” if you only want agendas for certain meetings (e.g. events whose title contains “sync”, “review”, or “1:1”). The matching-search route is the one I’d pick — you do not want an agenda auto-generated for every dentist appointment and focus block on your calendar.
2. Pull in context
This is the step that separates a useful agent from a generic one. The calendar event alone gives you a title, attendee list, and description. That’s often thin. Add one or two more sources:
- Previous notes: if your team keeps notes in Notion, Google Docs, or a Slack channel, add a “find” step that retrieves the last entry tagged with this meeting. Now the agenda can carry forward open action items.
- An intake form: a one-field Google Form (“What do you want covered Thursday?”) whose responses feed in. Cheap and shockingly effective for recurring team meetings.
If you skip this entirely, the agent still works — it’ll just generate a competent but generic structure from the title. That’s fine for a first version. Ship it, then add context.
3. Write the prompt
Add the AI step and treat the prompt like a brief, not a wish. Map the calendar fields into it. Something like:
- Role: “You are an executive assistant preparing a focused meeting agenda.”
- Inputs: insert the event title, attendees, description, and any retrieved notes as variables.
- Instructions: “Produce 3–5 agenda items. Each item: a clear topic, the desired outcome (decision / update / brainstorm), and a time box in minutes that fits the meeting length of
{{duration}}. Carry forward any unresolved action items from the prior notes. End with a one-line ‘pre-read’ if attendees should prepare anything. If the inputs are too vague to build a real agenda, say so and list the 2–3 questions the organizer should answer instead.” - Format: “Return clean Markdown. No preamble, no ‘Here is your agenda.'”
That last instruction in the role block — telling it to flag vague input instead of inventing filler — is what keeps the agent honest. Without it, an LLM will happily manufacture a polished agenda for a meeting it knows nothing about, and confident nonsense is worse than no agenda.
4. Deliver it
Pick one destination to start:
- Update the calendar event — append the agenda to the event description. Everyone who opens the invite sees it. Lowest friction.
- Post to Slack — message the channel or DM the organizer a few hours ahead. Best for visibility and quick edits.
- Email the attendees — formal, good for external or client meetings.
I’d start with Slack to the organizer only. That gives a human a chance to glance at the draft and fix it before it goes wide — which you want for the first few weeks until you trust the output.
The part most guides skip: keep a human in the loop (at first)
Fully autonomous “it posts straight to all attendees” is the goal, but earn it. For the first dozen meetings, route the draft to the organizer with a “looks good / edit” moment built in. In Slack that’s as simple as the agent posting the draft and the organizer copying it into the invite. After a couple of weeks you’ll know whether the prompt and context are solid enough to let it publish on its own. Skipping this trial period is the single most common way these agents embarrass someone.
When NOT to build this
A few honest cases where the agent is the wrong move:
- Ad-hoc meetings only. If your meetings aren’t recurring and have no prior context, the agent is just dressing up the title. A saved template you fill in by hand is faster.
- High-stakes external meetings. Board meetings, legal calls, key client pitches — write those agendas yourself. The downside of a sloppy auto-agenda outweighs the time saved.
- Sensitive content. If meeting context includes confidential data, check whether your AI step’s model retains or trains on inputs. Use a provider/setting with no data retention, or don’t feed it the sensitive bits.
- You have under ~5 meetings a week. The setup-and-maintenance time won’t pay back. Just use a checklist.
Sharpening it over time
Once the basic loop runs, the high-value upgrades are:
- Different templates per meeting type — a 1:1 agenda looks nothing like a sprint review. Branch on the event title and swap the prompt.
- Close the loop with notes — pair this with a second automation that captures action items after the meeting, so next time’s agenda starts with “open items from last week.”
- Let people reply — a quick form or Slack thread where attendees add topics before the agent drafts.
FAQ
Do I need to pay for an OpenAI or Anthropic API key?
Not necessarily. Zapier’s built-in “AI by Zapier” step runs without you supplying a key, so you can ship a first version paying only for your Zapier plan. The trade-off is less control over the model and prompt. In Make or n8n you’ll connect your own key — usually pennies per agenda at these volumes — which gives you better model choice and cheaper scaling.
How accurate are the agendas, really?
For recurring meetings with decent context fed in, genuinely good — often better-structured than what a rushed human writes at 8:59am. For meetings where you give it only a title, it produces a reasonable generic skeleton but can’t know specifics, so it may guess. That’s exactly why the prompt should instruct it to flag thin input rather than invent. Treat the output as a strong first draft, not gospel.
Will this work with Microsoft Teams and Outlook instead of Google?
Yes. Swap the Google Calendar trigger for the Outlook/Microsoft 365 calendar trigger and the Slack action for a Teams “send channel message” action. The middle (context + AI + prompt) is identical. All three platforms above support the Microsoft connectors.
Your next step
Don’t try to build the perfect version. Pick one recurring meeting you’re in this week — a weekly team sync is ideal — and build the four-step Zap for just that one: matching-search trigger, the prompt above, deliver to yourself in Slack. Run it for the next two meetings, tweak the prompt where the output missed, then add prior-notes context. Once one meeting is humming, cloning it to the rest takes minutes. The hard part was never the AI; it was deciding to stop writing agendas by hand.