Inbox Triage Agent: Claude+Make

A founder gets four hundred emails a day and the three that matter, an outage, a key customer about to churn, an investor reply, drown in the flood. Checking constantly is a tax on focus; not checking risks missing the one that mattered. Either way the inbox wins.

This build hands the watching to an agent. It reads every email as it arrives and pings the founder in Slack only when something is truly urgent by rules he sets, staying completely silent otherwise. He stops checking his inbox and starts trusting a tap on the shoulder when, and only when, it counts.

We wire it in Make, the last build in our AI agents set. The agent here only watches and flags, it never replies, so the risk is simply missing or over-flagging, which you tune. The skill is teaching it what “urgent” means for you, and trusting its silence.

What does this workflow actually do?

In one line: an agent watches your inbox and taps you only for the urgent. Make watches your mail. Claude reads each email against your definition of urgent and, only when it qualifies, posts an alert to Slack with the gist. Everything else, it stays silent. You stop monitoring and start trusting the tap.

A few real cases, none of them the usual ones:

• The four-hundred-emails founder above, finally free to ignore the inbox and trust the tap.
• an executive assistant filtering a principal’s inbox down to what truly needs them.
• a small-agency owner buried in cc’s who only wants the client fires.
• a freelancer separating client-urgent from the steady hum of noise.

The cost of a busy inbox is not the reading, it is the constant checking. An agent that watches for you, and stays quiet until it matters, gives the focus back.

What makes this an agent, and how do you keep it safe?

It is an agent because it decides what deserves your attention and acts, pinging you, on its own judgment. Its autonomy is judgment, not sending, so the failure modes are gentle: it might over-flag or miss one, both of which you tune.

The guardrails fit the job. A clear definition: you spell out exactly what counts as urgent, so it is not guessing at your priorities. Bias toward silence: you tell it that when in doubt it should stay quiet rather than cry wolf, because an alert channel only works if you trust it. Nothing is deleted or sent: every email stays in your inbox; the agent only adds a flag. Claude judges; Make watches and pings; your definition of urgent is what makes the silence trustworthy.

What do you need before you start?

• A Make account. The free 1,000 operations a month covers a heavy inbox.
• A Claude API key from the console, not the chat app. See how to use Claude.
• A Gmail account.
• A clear, written definition of what “urgent” means for you, and a Slack channel for the taps.

Two Make words. A scenario is the whole watch-and-flag automation. A module is one box. Three boxes, one agent.

How does the workflow work, step by step?

Three modules, left to right:

Module	App	What it does
1. Trigger	Gmail	Watches the inbox and fires on each email
2. Agent	Claude	Decides if it is urgent, by your rules
3. Output	Slack	Pings you only when it is

Step 1: Watch your inbox

Create a scenario and add Gmail, Watch emails, connect your account, and watch your inbox. This trigger hands the next step each email.

Step 2: Let the agent judge urgency

Add a Claude module, Create a Prompt, paste your key, pick Sonnet. Define urgent precisely, and bias it toward silence:

Decide if this email is URGENT for me. Urgent = an outage, a key
customer threatening to leave, a deadline today, or an investor
reply. Email: {{subject and body from Step 1}}

Reply "URGENT: " only if it clearly meets the bar.
Otherwise reply exactly "QUIET". When in doubt, choose QUIET.

The precise definition and the bias-to-QUIET are the guardrails: an alert channel that cries wolf gets muted, so silence is the safe default.

Step 3: Ping only when it matters

Add Slack, Send a Message behind a Make filter: if Claude’s reply starts with “URGENT:”, post the alert with the gist; if “QUIET”, do nothing. The email always stays in your inbox; the agent only decides whether to tap you.

How do you run it and check the result?

Click Run once on an urgent email and a routine one. When I tested this, the agent pinged on a fake outage and stayed silent on a newsletter, once I defined urgent tightly and told it to default to quiet. Tune the definition over a week until you trust both its pings and its silence.

After that, you can close your inbox and work. The agent watches all four hundred emails so you do not have to, and taps you only for the three that change your day.

What does this cost to run?

Piece	Free tier	If you outgrow it
Make	1,000 operations/month free	Core plan from about $9/month
Claude API	Pay per use	Judging an email on Sonnet costs a fraction of a cent
Gmail	Free	Free
Slack	Free plan works	Paid from about $7/user/month

Each email is three Make operations and a sliver of Claude. A heavy inbox can use operations, so point it at the inbox that matters. More on tiers in our Make guide.

What can go wrong, and how do you avoid it?

• It cries wolf. Bias it toward QUIET and tighten the definition; an alert channel only works if you trust it.
• It misses an urgent one. Widen the definition with the example it missed; tune over a couple of weeks.
• Operations burn on a huge inbox. Point it at a key inbox or label, not every address you own.
• You do not trust the silence. That trust is earned by tuning; until then, spot-check what it called QUIET.

The same watch-and-judge shape powers many agents. See Make AI scenarios.

How do you build this in Zapier or n8n instead?

The three jobs do not change, only the names do.

Job	Make	Zapier	n8n
Catch an email	Watch emails	New Email trigger	Gmail Trigger node
Judge urgency	Claude module	Claude (Anthropic) action	Anthropic node
Ping if urgent	Send a Message + filter	Slack + Filter	Slack node + IF

Make and Zapier both handle the flag-if-urgent filter cleanly. Zapier vs Make vs n8n compares all three.

Common questions

Will it delete or move my email?

No. It only reads and, when warranted, pings you in Slack. Every email stays exactly where it is.

How does it know what is urgent to me?

You define it in the prompt, an outage, a key customer, a deadline, an investor. The clearer your definition, the better it judges.

What if it over-alerts?

Bias it toward staying quiet and tighten the definition. An alert channel only works if you trust that a ping means something.

Can it ping me by text instead of Slack?

Yes, swap Slack for an SMS module. The agent logic is the same.

Is the chat app the same as the API?

Same models, different door. Make needs an API key from the console, not the chat login.

Sources and official docs

• Creating a scenario (Make Help)
• Claude API overview (Anthropic)
• Gmail Help Center (Google)
• Slack Help Center
• Email overload — Grokipedia

Last reviewed: May 2026. Make, Claude, Gmail, and Slack update their interfaces; check the official pages for exact button names.

You may also like

• AI Automation for Beginners (the hub)
• Build a Telegram auto-reply bot with Claude
• Draft support replies for an agent to review
• Make.com Complete Beginner’s Guide
• How to use Claude AI
• Make AI Scenarios: automations that save hours
• The AI Tools Directory