Agent Mail for ad ops: run tasks from your inbox without opening a dashboard
The gap between noticing something and acting on it is usually a login. Agent Mail closes it: send the agent a task from your phone, get a reply telling you exactly what it would run and where, confirm it, and read the finished report in the same thread. This is a walkthrough of that loop — plus the standing-rule pattern that handles the mail you already know is coming.
The workflow, start to finish
- 1Email the agent inbox from the address you sign in with — subject and body become the goal, exactly as you'd phrase it in the app.
- 2It replies with what it would run and in which organization. Nothing has started yet.
- 3Reply `confirm` (or 确认), or tap the link in that reply.
- 4The task runs and replies the report into the same thread. Anything that would change an account is waiting for approval, not already applied.
Where email genuinely beats the app
| Situation | Why email wins |
|---|---|
| You thought of it on your phone, away from a laptop | One email beats a mobile login and a dashboard you have to navigate |
| A client emails you a question about their account | Reply with `@agentplanners` and an instruction — the client's own email becomes the context |
| You want a report without babysitting the run | The reply lands in your thread when it's done; no tab left open |
| Something recurring already arrives by email | A standing rule handles it, and no one has to remember |
The @mention pattern for client threads
In an existing thread — a client asking why spend jumped, a colleague forwarding an alert — write `@agentplanners` followed by an instruction. The instruction becomes the task and the rest of the email becomes context, so you don't restate what's already written above. Quoted history and signatures are stripped first, so a long chain doesn't confuse the goal, and an explicit mention always beats a standing rule.
Standing rules for the mail you already expect
Some email is predictable: a weekly performance export, a budget-pacing alert, a client's Monday question. Author a rule in-app — conditions plus a standing instruction — and matching mail runs without the per-email confirmation, because setting the rule up while signed in *was* the consent.
- Conditions match on sender, recipient, subject or body. A rule with nothing filled in matches nothing, so you can't create a match-everything rule by accident.
- Each rule fires at most 10 times a day, and at most 3 rules apply to any one email.
- The task belongs to whoever wrote the rule.
- Replying back into the triggering thread is off by default — the sender might be a client, and a report in their thread would show them your data.
Useful things to send
- "Audit last week's Google Ads spend and flag anything wasting budget."
- "Which campaigns are pacing over budget this month?"
- "Compare Meta and Google cost per acquisition for the last 30 days."
- `status` — replies with your five most recent tasks and links, read-only.
- "Every Monday at 9am, send me last week's performance summary" — say the cadence and the agent sets up the schedule.
What it doesn't do
It doesn't apply changes because you confirmed. Confirming starts the task; every account change it proposes still queues behind the same approval gate as a task started in the app. It also won't act on mail from an address that isn't a member of your organization unless a rule you wrote covers it. Full mechanics and limits are in the email tasks documentation.
Frequently asked questions
- Can I run Google Ads tasks by email?
- Yes — email the agent inbox from your sign-in address with the task in the subject and body. It replies with what it would run and where, and starts only after you reply "confirm" or click the link in that reply.
- Do I need to install or configure anything?
- No. There are no credentials to set up, because the sending address is the identity — the email has to come from the address you sign in with, which is what links the task to your organization.
- Can I trigger a task from a client's email thread?
- Yes — write @agentplanners followed by an instruction in the thread. The instruction becomes the task and the surrounding email becomes context. Quoted history is stripped so a long chain doesn't muddle the goal.
- Will the agent reply into my client's thread with internal data?
- Not unless you deliberately turn that on for a specific rule. Replying into a triggering thread is off by default precisely because the sender may be an outsider.
- How many tasks can I start by email per day?
- 10 per sending address per 24 hours. Past that you get a rate-limit reply pointing you to the app.