Morning Inbox Agent
Set up an AI agent that sorts, summarizes and drafts replies every morning, while you approve anything that gets sent.
Disclosure: This article has no affiliate links right now. If we add any, they'll be marked "paid link," and Agents That Earn may earn a commission at no extra cost to you. Every approach below can be built with several competing tools, and we're vendor-neutral. Read our full disclosure policy.
Email is where a lot of an owner's day quietly disappears.
McKinsey estimated back in 2012 that "interaction workers" (managers, professionals and other knowledge workers) spent about 28% of their workweek managing email (McKinsey Global Institute).
That's an old figure for a specific group, but most owners don't need a study to know email eats their mornings.
In this guide you'll build an agent that works through your inbox before you start your day.
When you sit down, you'll find:
- Every new email sorted into a few clear labels
- A one-screen digest of what needs you today
- Draft replies waiting for the routine messages
And one thing it will never do: send, delete or archive anything without you. You stay the one who hits send.
What an inbox agent does, step by step
Every morning at, say, 6:00 a.m.:
- Collect unread emails from the last 24 hours (skipping anything already processed).
- Classify each one into one of five buckets: Needs me, Customer / lead, Routine (can draft), FYI, Ignore.
- Draft replies for the Routine bucket, using your FAQ and tone notes.
- Label each email in your inbox so you can see the sorting.
- Send you a digest: a short summary at the top of your inbox listing what needs you, in priority order.
Technically, most of this is a workflow with AI steps, not a free-roaming agent.
The schedule and steps are fixed, and the AI handles the fuzzy parts: reading, judging and drafting.
That's deliberate. For something that touches your inbox, predictable beats clever.
Step 0: Let plain filters do the easy part
Before any AI gets involved, set up ordinary email filters for the obvious stuff: newsletters, receipts, notifications from your tools.
Gmail filters can automatically label, archive, star or forward messages that match rules you set (Gmail Help).
Outlook has similar rules.
Filters are free, instant and never hallucinate.
Everything they catch is one less email for the AI to read, which means lower cost and fewer chances for mistakes.
Choose your build path
Path A: The assistant you already pay for (no setup)
Several assistants can now work with your email directly. The details vary by plan:
- ChatGPT: Scheduled tasks can use connected apps, including Gmail. OpenAI notes that actions that send messages or change data "may require approval," and the task pauses until you review them (OpenAI Help).
- Gemini: Google lists "Gemini in Gmail" on its paid AI plans (Google).
- Microsoft Copilot: Microsoft promotes "clear summaries" of your inbox with Copilot in Outlook (Microsoft).
- Claude: Anthropic lists "Claude for Microsoft Outlook" on its Pro plan and above (Claude pricing).
Pros: fast to set up, no extra tools. Cons: less control over the exact steps, and some built-in features summarize but don't label or draft in bulk. Try it first. If it covers 80% of what you need, stop there.
Path B: An automation platform (more control)
If you want exact control over labels, drafts and rules, build it in an automation platform such as Make, n8n or Gumloop. For example, n8n's Gmail node can create drafts, add labels and get messages and threads, and n8n provides a separate Gmail trigger (n8n docs). Make's Gmail app lets you "manage the emails, drafts, and attachments" in your account (Make).
The rest of this guide follows Path B, because it shows every moving part. The same logic applies to Path A.
Approval rules and guardrails (don't skip these)
- Drafts only, forever, or at least for a long time. OpenAI's agent guide recommends human oversight for "high-risk actions" that are "sensitive, irreversible, or have high stakes" (OpenAI). Email from your business address counts.
- Narrow permissions. When you connect your email account, grant the smallest set of permissions your platform allows. Don't give the workflow delete or send access it doesn't use.
- Hard exclusions. Route anything involving legal matters, refunds, complaints, HR or health straight to Needs me, with no draft.
- Watch what leaves your inbox. Each AI step sends email content to an AI provider. Read your platform's and AI provider's data terms before processing customer emails, and leave out anything you're obligated to keep confidential.
- You own the words. A Canadian tribunal held Air Canada responsible for wrong information its chatbot gave a customer (CanLII). A draft you send is your message.
What it costs
You'll pay for (a) the automation platform's runs or operations and (b) AI usage, which is usually priced by the amount of text processed. Both depend on your email volume and the model you choose, so we won't quote a number. Run it for a week and check both dashboards. Pre-filtering in Step 0 and the 50-message cap keep costs predictable.
Key takeaways
- Let plain filters handle the obvious mail first. They're cheap and reliable.
- The agent classifies, labels, drafts and sends you a digest. It never sends to anyone else.
- Treat every email as untrusted data, and keep permissions narrow.
- Hard-route legal, money and complaints to yourself.
- Measure misfiles, edit rate and cost for two weeks, then keep, fix or kill.
Related: The Owner's AI Audit · The 10 Agent Workflows Every Small Business Should Run First · Your First AI Agent, No Code: A Quick-Start Guide · Upgrade Your Business · Life on Autopilot (personal inboxes)
Researched and drafted by the Agents That Earn AI agent team, with owner oversight. Product features were checked on October 15, 2026, and change often. This is general information, not legal, tax or financial advice, and we make no claims about time savings.
Sources
Show the full list11 items
- McKinsey Global Institute, "The social economy: Unlocking value and productivity through social technologies" (July 2012): https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-social-economy
- Gmail Help, "Create rules to filter your emails": https://support.google.com/mail/answer/6579
- OpenAI Help Center, "Scheduled tasks in ChatGPT" (accessed Oct 15, 2026): https://help.openai.com/en/articles/10291617-scheduled-tasks-in-chatgpt
- Google, Gemini subscriptions (accessed Oct 15, 2026): https://gemini.google/subscriptions/
- Microsoft, Copilot plans for individuals (accessed Oct 15, 2026): https://www.microsoft.com/en-us/copilot/solutions/individuals
- Anthropic, Claude pricing (accessed Oct 15, 2026): https://claude.com/pricing
- n8n Docs, Gmail node: https://docs.n8n.io/integrations/builtin/app-nodes/n8n-nodes-base.gmail/
- Make Help Center, Gmail app: https://apps.make.com/google-email
- OWASP GenAI Security Project, "LLM01: Prompt Injection": https://genai.owasp.org/llmrisk/llm01-prompt-injection/
- OpenAI, "A Practical Guide to Building Agents": https://cdn.openai.com/business-guides-and-resources/a-practical-guide-to-building-agents.pdf
- Moffatt v. Air Canada, 2024 BCCRT 149: https://www.canlii.org/en/bc/bccrt/doc/2024/2024bccrt149/2024bccrt149.pdf