The weekly team meeting, the biweekly project sync, the monthly leadership review: these standing meetings are where a lot of work is steered, and where the record is usually worst. Someone takes notes one week, someone else the next, the format drifts, and three months in nobody can answer "wait, what did we decide about that?" because the decision is buried in a doc nobody kept consistently. An AI agent can fix the record without adding a note-taker: produce the same clean summary after every session and, crucially, carry the open threads forward so the series tells a continuous story.

This guide covers using an agent for recurring meeting summaries in five steps. It builds on how to set up your first AI agent, and it complements the messaging side of meetings in AI agent for meeting follow-ups.

What the agent does

A recurring meeting summary agent reads the record of each session, your notes or a transcript, and produces a summary in a fixed format: what was decided, what was updated, what is still open, and what comes next. Because it runs on the same meeting week after week, it does two things a one-off summarizer cannot. It keeps the structure identical so any session is instantly readable, and it remembers the prior sessions so it can show what moved. It drafts the summary; a person gives it a quick check before it is shared.

This fits an agent well because summarizing to a fixed template is a bounded, repeatable task, and the inputs and the desired output are both clear (Anthropic, "Building Effective Agents", 2024). The format is set, the meeting recurs, and the agent applies the same lens each time. The judgment, whether a summary captured the real significance of a discussion, stays with a participant who reviews it.

Why a person stays in the loop

A meeting summary becomes the shared memory of what was agreed, so accuracy matters and a person should confirm it. The agent's job ends at a drafted summary; a participant checks it. If you are weighing whether an agent suits this connected, context-carrying task over a plain transcription tool, what is an AI agent explains the difference, because the continuity across sessions is exactly what makes this agent-shaped rather than a one-shot tool.

Why recurring meetings differ

Summarizing a one-off meeting is a self-contained job: read this, condense it, done. A recurring meeting is different because the value is in the series, not the single session. The point of a weekly sync summary is not just "what happened this week"; it is "what changed since last week, what we decided that is now settled, and what is still hanging." A summary that ignores the prior weeks misses the story the meeting actually tells.

That is why a recurring-meeting agent earns its keep. It treats each session as the next chapter, not a standalone document. It knows the decision made two weeks ago, notices that an open item finally closed, and flags a thread that has been unresolved for a month. If your need is instead to track who committed to what across all meetings, that specific job is covered in AI agent for action item tracking, and updating a deal record from a sales call is covered in AI agent for meeting notes to CRM.

1. Define the outcome

Write the result in one sentence first. For example: "After each weekly project sync, a summary in our standard format, decisions, updates by workstream, open items with owners, and changes since last week, ready for the lead to skim and post." That sentence fixes the format, the continuity requirement, and the human check at the end.

Why outcome-first matters here

Stating the outcome locks the format, which is the whole point of a recurring summary. Describe the result and the structure follows: the sections, the carry-forward of open items, and who reviews. This is the describe-the-result approach the platform is built on, set out in how to set up your first AI agent. You describe the summary shape you want once; the agent reproduces it every session.

2. Connect access

The agent needs to read the source for each session, your notes tool, a shared doc, or a meeting transcript, and a place to keep the running record of prior summaries so it can carry threads forward. It needs to draft a summary for review. It does not need to act on anything beyond producing the document. Grant read access plus the ability to write the draft summary, and keep distribution under a person's control.

Scope access tightly

Give the narrowest access that does the job and review what the agent can see before connecting it, since meeting notes can hold sensitive discussion. The agent reads and drafts; it takes no other action, which keeps it low-risk by design. Treat the notes with the care any internal data deserves, as the broader AI agent security best practices guide explains.

3. Summarize consistently

With access in place, the agent reads the session and writes it into your fixed format. Every summary has the same sections in the same order, so a reader who has seen one has seen them all and can jump straight to the part they care about. Consistency is not cosmetic here; it is what lets a busy team actually use the summaries rather than skim and forget them.

read_session(notes|transcript) -> this week's raw record
apply_template(format)         -> fixed sections, same order
extract_decisions(session)     -> what was settled
extract_open(session)          -> what is still unresolved

Holding the format steady across sessions is what turns a pile of meeting notes into a usable archive. Search for a decision and you know which section it lives in. Compare two weeks and the structure lines up. The agent's discipline about format does the quiet work that human note-takers, rotating and rushed, rarely sustain.

4. Track decisions across sessions

Now the agent adds the continuity that makes a recurring summary worth more than a stack of one-offs. It carries forward the open items from last session, marks which closed and which are still pending, notes decisions made and whether anything reversed one, and surfaces threads that have lingered too long. Each summary therefore shows not just this session but its place in the arc of the meeting.

What good continuity reads like

Good continuity is explicit about change. "Decided: ship date moved to Sep 1 (was Aug 15), per this week's capacity review." Or: "Still open after 3 weeks: pricing sign-off, owner Finance, no movement." That tells a reader what is settled, what shifted, and what is stuck, at a glance. A summary that just lists this week's discussion, with no memory, leaves them to reconstruct the history. The agent maintains the thread; a person confirms it is right. If you run many recurring meetings and want to size the load first, how to estimate agent cost before deploying shows how.

5. Distribute

The final stage gets the checked summary to the people who need it: posted in the team channel, added to the meeting doc, or sent to attendees. A participant gives it the quick review, corrects anything off, and shares it. The agent keeps its copy in the running record so the next session has the context it needs. A summary that lands reliably, in the same place, in the same shape, is one people come to trust and actually read.

A summary records, it does not decide

Worth stating plainly: a recurring summary is a record of what people decided, not a decision-maker. The agent captures and carries the decisions; it does not make them or judge whether they were wise. Treat the summary as a faithful, continuous account that a participant has confirmed, and it becomes the shared memory the meeting needs, without pretending to a role it should not have. For a daily, cross-tool version of this rollup habit, see AI agent for Notion daily rollup.

The Gravity way to run it

On a platform like Gravity you do not build any of this. You describe the outcome, "after each weekly sync, summarize it in our format and carry forward the open items and decisions, ready for me to check," and an expert-built agent handles the reading, the consistent formatting, and the continuity, then hands back the summary in about 60 seconds. You pay only when it runs, at $1 for 1,000 credits.

Frequently asked questions

Can an AI agent summarize my recurring meetings?

Yes. An AI agent can read the notes or transcript from each session of a standing meeting and produce a summary in the same format every time, covering decisions, updates, and open items. It drafts the summary for a quick human check before it goes out, so the record stays accurate.

How is this different from a one-off meeting summary?

A one-off summary stands alone. A recurring summary adds two things: a consistent format across every session so they are easy to scan and compare, and continuity, the agent carries forward open threads and decisions from prior sessions so the summary shows progress over time, not just what happened once.

Does the agent track decisions over time?

Yes. Across a recurring meeting it maintains a running record of decisions made and threads still open, so each new summary notes what was decided, what changed since last time, and what remains unresolved. That continuity is the main reason a recurring-meeting agent beats summarizing each session in isolation.

Is it safe to connect an AI agent to my meeting notes?

It can be, with tight scope. The agent needs read access to the notes or transcript and the ability to draft a summary for review. It does not need to act on anything. Grant the narrowest permission that does the job, review what it can see, and keep a person on the check before a summary is shared.

How do I set up a recurring meeting summary agent?

Define the outcome first: a consistent summary after each session, carrying forward open items, ready for a quick check. Connect read access to your notes source, set the format, and route the draft for review. On a platform like Gravity you describe the outcome and an expert-built agent produces the summary in about 60 seconds.

Three takeaways before you close this tab

Sources