Hand ChatGPT Your Recurring Work: Team Tasks That Run While You Sleep

What OpenAI launched at DevDay 2026

On September 29, 2026, at its DevDay event in San Francisco, OpenAI announced a batch of features that turn ChatGPT from a chatbot you visit into a workspace where work happens without you. The headliner was Dots, always-on agents with their own cloud computer. But the feature that matters most for a small business is quieter: Teams and Team Tasks.

A Team Task is a job you define once and that ChatGPT then runs on its own — either on a schedule (every weekday morning, every Friday afternoon) or when something happens (a new email arrives, a message lands in Slack). According to OpenAI's developer-community recap of the announcements, teams can now "assign recurring tasks as a team," and ChatGPT responds to triggers such as a new email or Slack message. Several independent write-ups of the event — including Mad Robot's full DevDay recap and UC Today's breakdown — confirm the same two mechanics: schedule or event trigger, running in the cloud using the team's connected accounts.

The same announcement put ChatGPT itself inside Slack and Microsoft Teams: you mention @ChatGPT in a channel or thread, and colleagues can join in without their own ChatGPT licence. And a Meetings plugin, in beta on the Mac desktop app for Pro and Business users, takes notes, saves action items into the shared workspace, and deletes the audio once notes are ready — The Decoder's coverage, summarized by WindowsForum, notes the audio deletion as a deliberate privacy choice.

Two honest caveats before we go further. First, Teams and Team Tasks are currently announced for ChatGPT Business and Enterprise workspaces — not for individual Plus accounts. Second, this is week-one news: coverage from AIO Copilot and BenchLM agrees that details like exactly which connectors support event triggers are still being filled in. Treat this as a direction you can start preparing for, not a finished product you buy tomorrow.

Why recurring work is the best first thing to automate

Most AI advice asks you to automate a whole job. That is the wrong starting point. The work that actually eats a small business owner's week is not dramatic — it is the same handful of jobs, repeated endlessly:

• Checking what came in overnight and deciding what needs attention today.
• Following up on enquiries that arrived by email and have not been answered.
• Scanning pending quotations at the end of the week to see which ones need a nudge.
• Preparing a simple summary of the week for yourself or a partner.

These jobs share three properties that make them ideal for automation. They recur on a predictable rhythm, so a schedule fits them perfectly. They have a fixed format, so you can describe the expected output once. And they are low-stakes — a draft briefing or a reminder list can be checked in two minutes, so a mistake costs you almost nothing.

That is exactly what a Team Task is built for. You write down what should happen and when, once, and the team shares the result. OpenAI's own rollout guidance, published through its Champion resources, says the same thing in plainer business language: start with one workflow, choose a recurring task your team already understands, and define the expected result so someone can review it.

If your team already wastes hours re-typing context into chatbots, this is the next step up from Gemini Skills, which we covered as a way to stop re-typing instructions — and it complements the broader recurring routines like the five mundane workflows most SMBs can automate this quarter.

The practical use case: three Team Tasks in one sitting

The best way to understand Team Tasks is to set three of them up. Below is a follow-along sequence you can do in one sitting with a ChatGPT Business workspace. Each recipe gives you the exact instruction text to paste into the task — copy it, adjust the bracketed bits to your business, and the task will run on its own from there. Where natural, this is the kind of recurring workflow automation a team refines over time rather than building once.

Step 1: Make sure you have a Team

Team Tasks live inside a ChatGPT Team. In the ChatGPT web app, open the workspace menu (top-left) and create a Team if you do not have one, then invite the one or two people who will check the task outputs — a partner, a supervisor, or yourself on two accounts.

Expected result: you see a shared Team space in the sidebar. If you are on a plan without Teams yet, you can still read the rest of this post as a preparation guide — the recipes below are exactly what you will paste in when access arrives.

Step 2: Recipe 1 — the daily morning briefing

Create a new Team Task. Set the schedule to every weekday at 8:30 AM. Connect the email account that receives your business enquiries. Paste this as the task instructions:

Every weekday at 8:30 AM, prepare my morning briefing.

Look at all business emails that arrived in the last 24 hours in the connected inbox.
For each one, decide which of these it is:
1. A new sales enquiry — someone asking about our products, services, prices, or availability.
2. A follow-up on an existing order or quotation.
3. A bill, receipt, or payment notice from a supplier.
4. Spam, newsletters, or promotions — ignore these completely.

Then write the briefing in this exact format:

TODAY'S BRIEFING — <date>
New enquiries: <number>. List each with the sender name and one line on what they asked.
Orders needing attention: <number>. List each with what is expected next.
Bills to pay: <number>. List each with amount and due date if shown.
Suggested first action: <one sentence saying what I should do first this morning>.

Rules: keep the whole briefing under 250 words. Never mark an email as read or reply to anything. If an email is unclear, say "needs a human look" next to it.

Expected result: the next weekday morning, your team finds a short briefing in the shared space — a numbered list of what arrived and one suggested first action. It takes two minutes to read instead of twenty minutes of inbox scrolling.

Step 3: Recipe 2 — the new-enquiry draft reply

Create a second Team Task. Set the trigger to "when a new email arrives" in the same inbox, with a filter for emails that look like sales enquiries. Paste this:

When a new sales enquiry email arrives, draft a reply for my review.

Read the enquiry carefully and draft a reply that:
1. Greets the sender by name.
2. Repeats back their requirement in one sentence, so they feel heard.
3. Answers their question from what you know about our business (see the background below).
4. Asks exactly one follow-up question that helps move the sale forward — for example their timeline, quantity, or budget range.
5. Closes with my name and phone number.

Background about our business: [paste 3-5 lines — what you sell, your town/area of service, your working hours, and one line on why customers pick you].

Rules: write in simple, polite English. Keep the draft under 150 words. Do NOT send the reply — save it as a draft for me to review and send. If the enquiry is in another language, draft the reply in that language too.

Expected result: within minutes of an enquiry landing, a draft reply waits in your team space — you read it, adjust a word or two, and send. Nobody types the same polite greeting a hundred times again.

Step 4: Recipe 3 — the Friday pending-quotation sweep

Create a third Team Task. Set the schedule to every Friday at 4:00 PM. Paste this:

Every Friday at 4:00 PM, sweep my pending quotations.

Look through this week's sent emails for quotations or price quotes I sent out.
For each quotation sent more than 3 days ago with no reply, prepare a follow-up note:
- Customer name and what the quote was for.
- Date the quotation was sent.
- A two-line follow-up message I could send, referencing the quotation and asking if they need any clarification.

Present the result as a simple table with these columns: Customer | Quotation | Sent on | Follow-up message.

Rules: do not contact any customer. This is a preparation list only. If you cannot find the date a quotation was sent, mark it "date unclear".

Expected result: every Friday, a table of quotations that went quiet — with follow-up lines ready to copy. Stale quotes stop being forgotten, and follow-ups stop depending on memory.

Step 5: Review the first runs together

Team Tasks can be refined by the team — BenchLM's DevDay notes specifically call out that teams "refine the instructions together." For the first week, read each task's output and note what to change: too long, missing a detail, misreading a category. Edit the task instructions directly — the recipe text above is your starting draft, not a finished product.

Expected result: after a week of small edits, the tasks need almost no corrections. That is when you stop reviewing every run and start skimming.

Step 6 (optional): bring ChatGPT into your chat app

If your team lives in Slack or Microsoft Teams, connect @ChatGPT to a channel. From then on, anyone can mention it in a thread to ask about the shared work — no extra ChatGPT licences needed for colleagues to join in. The practical payoff: the person who checks the morning briefing can ask follow-up questions right where the team already talks.

Expected result: fewer screenshots and forwarded emails — the task outputs and the conversation about them live in the same place.

Check the first runs like you'd check a new employee

A Team Task is a new team member who works while you sleep. Treat its first month the way you'd treat a new hire's:

• Give it bounded work. The three recipes above produce lists and drafts — things a human reviews in minutes. Do not give a week-one task the authority to send messages, pay bills, or delete anything.
• Define "done" before it starts. Each recipe above specifies an exact output format. A task with a vague instruction ("keep an eye on my inbox") produces vague output.
• Keep the human approval on anything outward-facing. The enquiry-draft recipe saves drafts for review rather than sending. That is not caution for its own sake — it is how you catch misread enquiries before a customer sees them.
• Track what it saves you. Once a month, compare: how long did the morning briefing take before versus now? OpenAI's own rollout guidance suggests tracking corrections, review time, and usage — not because the numbers are exciting, but because they tell you which task to build next and which to retire.

One more sanity check, from the coverage that flags it most clearly: event-driven automations that act while nobody is watching are exactly what deserves a review habit. Start with read-only and draft-only tasks; widen the permissions only when the review record earns it.

The honest limits, in plain words

A fair post says what the new thing cannot do yet:

• Availability is narrow right now. Teams and Team Tasks are announced for Business and Enterprise workspaces. If you run your shop on a personal ChatGPT account, you cannot set these up today — though the recipes above will still apply when access widens.
• Event triggers depend on connectors. A task that fires "when a new email arrives" needs your email connected to the workspace, and the coverage so far does not spell out every connector that supports events. Test each trigger and watch the first few runs.
• The Meetings plugin is a beta on Mac only. Meeting notes that delete their own audio are a genuinely good privacy design, but the plugin is currently in beta on the macOS desktop app — Windows users wait, per The Decoder's reporting.
• A task is only as good as its instructions. Vague instructions produce vague, confident output. The weekly refinement habit in Step 5 is the whole game.

None of this is a reason to ignore Team Tasks. It is the reason to start with the three low-stakes recipes above, review them for a month, and let the evidence — not the launch hype — decide what you automate next.