Office & managementNewMethod A proven way to do the task; no measured customer result for this exact workflow yet.
An AI agent that turns your emails and messages into a live, prioritised to-do list and day plan
Added 2026-10-14 · Last updated 2026-10-14
Role
Office & management · Typical users: managers, engineers, buyers, anyone whose work arrives by email, Teams/Slack and project tools
The problem
In a plant, the work you are asked to do rarely arrives as a task. It is buried in a supplier email ('can you confirm by Thursday?'), a Teams message from production, a comment in Asana or ClickUp, or the fourth reply of a long thread. Nobody re-reads all of it, so requests slip, the urgent item is found at 4 p.m., and the day is planned around whatever landed last. Microsoft's own field data shows the size of the problem: in one large study the average worker spent 11.7 hours a week reading and answering email.
How AI helps
An agent with read-only access to your mailbox, chat and work tools wakes up every 15 minutes, reads only what is new since its last run, and pulls out every request, commitment and deadline hidden in the messages. It merges duplicates (the same ask in an email and in Teams), scores each task P1/P2/P3 with clear rules (deadline, sender, impact, blocking others), keeps one live to-do list, and proposes a day plan that it re-shuffles when a new P1 arrives. You confirm, edit or dismiss; it never replies or acts for you.
Step by step
- Connect the sources with read-only permissions: mailbox (Outlook or Gmail), Teams or Slack (your DMs and a few key channels), and your work tools (Asana, ClickUp, Planner). Use your company's approved connector or tenant app, and write down exactly which scopes the agent has.
- Schedule a run every 15 minutes during work hours, with a 'since last run' filter: store the timestamp (or message IDs) of the last run and fetch only newer messages and comments. This keeps token use low; skip newsletters, notifications and automated mail with simple rules before any AI call.
- Extract tasks: for each new message, the AI lists every request made to you, every commitment you made ('I'll send it Friday'), deadlines, who asked, and a link back to the source message. Nothing found = nothing added.
- Deduplicate and merge: match new items against the open list (same requester + same topic, or same thread). Update the existing task (new deadline, new details) instead of creating a copy, and close tasks when a later message shows they are done.
- Score P1/P2/P3 with written rules. P1: due today or tomorrow, blocks someone else (production, a shipment, a customer), or safety/quality/customer-escalation. P2: due this week, or a request from your manager or a key customer/supplier with no hard date. P3: no deadline, information-only follow-ups, nice-to-have. Keep a short VIP sender list and let the rules beat the AI's guess.
- Build the day plan: fit P1s first around your calendar meetings, then P2s, with realistic durations and a buffer; list P3s as 'if time allows'. When a new P1 arrives, the agent proposes a revised plan and says what moves.
- Human confirmation: new tasks land as 'proposed'. You accept, edit or dismiss them in one view (Planner, Google Tasks, ClickUp or a daily digest). The agent never replies, forwards, accepts meetings or changes anything in the source systems.
- End-of-day review (5 minutes): the agent lists what was done, what slipped, open P1s for tomorrow, and false positives you dismissed. Use the dismissed items to tighten the rules and the prompt every week.
Tools
- Microsoft 365 Copilot (Outlook, Teams) + Microsoft Planner / To Do; Copilot Studio to build a scheduled agent
- Gmail + Google Tasks / Google Calendar (Gemini in Workspace, or Apps Script on a time trigger)
- Zapier, Make or n8n: scheduled 15-minute polling, connectors to Outlook, Gmail, Teams, Slack, Asana, ClickUp, and an LLM step
- ClickUp Brain / ClickUp Super Agents (email-to-task, used at VanHomes and Bell Direct)
- Asana AI (task extraction and prioritisation inside Asana)
- Motion and Reclaim.ai: turn the prioritised list into time blocks on your calendar
- Any general LLM (ChatGPT, Claude, Copilot, Gemini) with the prompt below to test the logic on a pasted batch first
Result
Method case: no published figure measures this exact agent end to end. Related published figures: (1) Microsoft researchers, randomised study of about 6,000 workers at 56 firms ("Early Impacts of M365 Copilot", April 2025): "Workers with access to Copilot reduced this time by 7% on average (12 minutes per week)" of email reading, and "Treated workers who used Copilot at least once a week reduced their weekly time reading emails by half an hour (18%)." (2) Microsoft researchers, 7,137 workers ("Shifting Work Patterns with Generative AI", 2025): "workers with Copilot reduced their time using Outlook for email by 1.4 hours per week, a 12% reduction". (3) ClickUp customer stories (vendor-reported): VanHomes (modular home manufacturer), where "AI takes customer emails, breaks them into subtasks, and categorizes each issue automatically" and "it saves our team at least 10–15% of their time"; Bell Direct (online broker, not a manufacturer), where an agent triages "800+ emails" a day, with "2 FTEs' work handled by Super Agents" and a "20% boost in operational efficiency". Practitioner note (Frédéric D'Amours): I run this kind of agent for my own work. Every 15 minutes it reads what's new in my mail and messages, adds the tasks it finds to my list with a P1, P2 or P3, and updates my plan for the day. I don't have a measured before/after figure; the gain I notice is that requests no longer get lost in threads and I start each day with a plan rather than an inbox.
Human review: cautions
- Give the agent read-only access. It must never reply, forward, accept invitations, or change tasks in the source systems without you clicking 'accept'.
- Mailboxes and chats hold confidential data (HR, pricing, customer drawings, personal info). Use only tools and connectors approved by your IT, keep the data in your company tenant, and do not paste mail into a public chatbot.
- Expect false positives (FYIs read as tasks) and misses (a request phrased as a question). Keep the human confirmation step, and check the dismissed list weekly to tune the rules.
- Never let it auto-reply or make commitments in your name; a wrong 'yes' to a customer or supplier costs more than the time saved.
- Watch the token cost: poll every 15 minutes, not every minute; fetch only messages since the last run; filter out automated mail before the AI call; send short excerpts, not whole threads. Check the usage bill after the first week.
- Published time-saving figures come from Microsoft researchers studying their own product and from ClickUp customer stories. Measure your own before/after (time in the inbox, missed requests) for two weeks.
Try it yourself
You are my executive assistant. Below is a batch of new messages since your last run (emails, Teams/Slack messages, Asana/ClickUp comments). Today is [date], my working hours are [8:00–17:00], and my meetings today are: [paste calendar]. My current open task list (may be empty): [paste: task | requester | due | priority] My VIP senders: [manager, key customers, key suppliers, plant manager] New messages (for each: source | from | date/time | subject or channel | text): [paste messages] Do this: 1. Extract every task: a request made to me, a commitment I made, or a deadline I must meet. Ignore newsletters, FYIs and automated notifications. For each task give: task (start with a verb) | requester | source + message reference | due date (only if stated or clearly implied; else 'none') | estimated duration | blocks someone? (yes/no, who). 2. Deduplicate against my open list: mark each item NEW, UPDATE (to which existing task) or DONE (a message shows it is finished). 3. Prioritise with these rules, and give the rule used: - P1: due today/tomorrow, OR blocks production, a shipment, a customer or a colleague, OR safety/quality/customer escalation. - P2: due this week, OR from a VIP sender with no hard date. - P3: no deadline, low impact, nice-to-have. 4. Propose a day plan: time blocks around my meetings, P1 first, then P2, with a 15% buffer; list P3 as 'if time allows'. Say what changed versus my previous plan and why. 5. List anything ambiguous as a question for me (unclear owner, unclear deadline), and any item you were unsure was a task at all. Rules: use only the messages given; never invent deadlines or requests. Do not draft or send replies. Output tables I can paste into my task tool. Use only the data I paste in; if something is missing, write 'not found' or list it as a question instead of guessing. Do not send, order, post or change anything in any system: produce drafts for my review and approval.
Want more like this? Join the newsletter
Sources
- https://arxiv.org/abs/2504.11443
- https://arxiv.org/abs/2504.11436
- https://clickup.com/customers/vanhomes
- https://clickup.com/customers/bell-direct
Weekly: 1–2 practical AI use cases for your plant
Free, one email a week. By subscribing you agree to receive the weekly newsletter (and its sponsor message) from AI Agents for Manufacturing. Every email includes an unsubscribe link, and you can unsubscribe anytime. See our Privacy Policy.