Stephen Turner
Akash Kumar
Jordan Lee
Taylor Scott
Use an AI agent to draft post-meeting follow-up emails
A follow-up email agent reads a finished meeting's notes and transcript. It drafts the email to attendees: what was agreed, who owes what and what happens next. The useful version drafts and waits. A person checks the facts, tone and recipients, then sends it.
The job: a draft, not a sent email
Give the agent one job. After a meeting with outside attendees, it drafts a follow-up email from what was actually said. You can check every sentence against the transcript, and nothing reaches a customer until a person decides it should.
Sending automatically is a bigger risk. A follow-up restates commitments in writing, often to people outside your company. A wrong date, an invented promise or the wrong recipient is hard to take back. Start with drafts, and only reconsider after you've checked enough of them to trust the pattern.
Inputs, output and what to leave out
If a commitment was tentative in the meeting, it should stay tentative in the draft. “Sam will look into the export issue and confirm a date” is accurate. “Sam will fix the export issue by Friday” is a new promise.
| Part | What it should be |
|---|---|
| Input | The completed meeting's notes, action items and transcript; optionally the attendee list |
| Output | Subject line, a two-sentence recap, decisions, action items with owner and date, next step |
| Must cite | The transcript moment behind each commitment, so the reviewer can check it |
| Must not | Add commitments, prices or dates that weren't stated; include internal-only discussion |
| Human step | Check facts, recipients and tone; then send from your own email |
Instructions you can adapt
This is a starting point, not a built-in template. Test it on meetings you remember well first.
- After each meeting with attendees from outside our organization, draft a follow-up email to them. Don't send it.
- Use only the meeting notes and transcript. If a decision, owner or date is unclear, write it as an open question.
- List each action item with its owner and any explicitly agreed date. Note the speaker and rough time it came from.
- Leave out discussion held without the external attendees, unoffered pricing and anything marked confidential.
- Keep the email under 200 words, in plain sentences, ending with the agreed next step.
Set it up in Bridge
Custom agents need a paid Bridge plan. You create one in four steps: Identity (name and instructions), Abilities, Tools and Triggers. Every agent can read transcripts.
Gmail is connected once, in Account settings → Integrations. The agent's Tools step then decides whether this agent may use it. With Gmail, an agent can draft replies and send follow-ups for you, within the access you already have.
That makes the tool choice the real control. Telling the agent “don't send” is a request, not a permission limit.
The After every meeting trigger runs the agent on the recap and transcript as soon as a meeting ends. Post-meeting follow-ups and email automation are among the jobs it's built for. Turn it on only once you trust the drafts and have decided whether the agent may use Gmail.
Ways to keep it draft-only
- Leave Gmail off for this agent and @mention it in a completed meeting's Assistant. The draft appears in the chat for you to copy.
- For a one-off, use the preset Gmail action in the recap's Assistant to draft a follow-up from that meeting.
Check the first drafts against the meeting
For the first ten or so meetings, read each draft beside the transcript. Include at least one meeting where nothing was agreed, to see whether the agent invents next steps. Fix the instructions for each problem before using the agent more widely. Look for three failures:
- An action item turned into a firmer promise.
- An owner assigned to the wrong speaker.
- Internal remarks leaking into an external email.
Sources & further reading
Keep readingUse an AI agent to review meeting action items →