Skip to content
OperatorNest

Work that keeps going after you close the chat

A background AI agent keeps working on a task after you close the chat or turn off your device. Send your operator a fare comparison across three weekends, close the app, and it works through each date combination in its own cloud workspace, then posts the result, or one question, in the channel you asked from.

How the work moves

  1. Starts when

    A task you send that will take longer than a few minutes to finish.

  2. Uses

    Its own cloud workspace, a browser, and any files or tools that task needs.

  3. Asks you before

    Anything that would send, pay, book, delete or publish before the task finishes.

  4. You get

    The finished result waiting in your channel, or a decision if one came up.

  5. On the record

    What it did, in what order, and how long each step took, for that task.

On this page

Work that doesn’t stop when you look away

OperatorNest is a background AI agent: once you hand off a task, it keeps working after you’ve stopped watching. Close the chat, turn off your laptop, and come back to a finished result.

Use it for work that takes longer than a conversation: research across a dozen sources, comparing options across several sites, or drafting a long document from scattered notes. You don’t keep a tab open or check in every few minutes for that work to progress.

How a background task runs

When a task will take a while, your operator moves it into its own cloud workspace and works through it step by step, whether or not your chat is open. For example:

“Compare business-class fares to Tokyo across the next three weekends, factoring in my usual airline preferences, and put together a short comparison.”

That can mean checking a dozen date and airline combinations, cross-referencing your past trips and building a table, which takes anywhere from twenty minutes to two hours depending on how much there is to check. You get the finished comparison, or a question if it hits one it can’t resolve on its own.

What you see while it’s running

Ask your operator for a status update at any point and it tells you where things stand, for example “checked 8 of 12 date combinations, no major fare drops yet.” If the task needs a decision partway through, such as which of two similar hotels to price out further, it asks and keeps working on anything that doesn’t depend on your answer.

When it finishes, the result lands in the channel you sent the task from, so there’s no separate app to remember to reopen.

Where background tasks show up

Any task bigger than a quick answer runs in the background automatically, with no mode to switch on. That covers a research request spanning dozens of pages, a multi-step web task like comparing vendors or filling out a long form, and a substantial draft like a report or a set of meeting notes. See web tasks and research assistant for two common shapes.

The same mechanism lets a scheduled task run at 3 a.m. without you present; see always-on agent for how scheduling and background work fit together. For a vendor-by-vendor view of schedule controls and failure details, read the scheduled AI tasks comparison.

Background tasks versus a schedule

A recurring task runs the same request on a cadence you set, like a Monday brief. A background task is a single piece of work too big for one sitting, like a deep comparison or a long draft, and it runs once, start to finish, however long that takes.

The two often work together. A weekly competitor brief is a recurring task; the research behind each Monday’s run happens in the background, like any one-off research request.

Control without babysitting

Background work follows the same rules as anything else your operator does. It stops for your approval before it sends, pays, books, deletes or publishes anything; see approvals. You can cancel a running background task at any time, and nothing it hasn’t already completed will continue.

If a task stalls, for example because a source it needs is unreachable, it tells you what got done, what didn’t, and why.

The record when it’s done

Every background task logs what it did and in what order: which sources it checked, what it ruled out, what it drafted, and how long each part took. Check it before you act on a fare comparison or a research summary. See receipts for what that record includes.

Example run

Overnight press list

25 targets found, a verified list, 11 pitches sent by mid-morning

Illustrative: this shows how the flow works, not a captured screenshot of a live task. Rows are described generically; no real journalist is named.

Hand-off · Email, 22:50

Subject: Overnight: press and podcast list. Find 25 journalists, newsletter writers and podcast hosts who covered personal AI agents, Meta Muse, Grok Bot or AI assistants in the last 90 days. For each, verify on their own site or recent article: name, outlet, the specific piece, and the best public contact route. Put it in a spreadsheet. Then draft a tailored pitch for each that references their piece. Use my Claude key for the drafts. Hold everything for my review in the morning; send nothing.

Brief

25 targets, 22 verified, 3 dropped, 25 drafts waiting

Found 25 candidates who covered personal AI agents, Meta Muse, Grok Bot or AI assistants in the last 90 days. 22 verified against their own site or byline: for example, a reporter at a national tech outlet who covered the Meta Muse launch on 8 September; a newsletter writer who ran a piece comparing always-on AI agents in mid-September; a podcast host whose recent episode covered agent security concerns. 3 dropped: one had no outlet affiliation that could be verified, one had no public contact route beyond a general inbox already at capacity per their bio, one credited byline turned out to be a syndication, not the original piece. All 25 have a tailored draft waiting, written in your tone and referencing their specific piece.

Approval

Send the 11 approved pitches (8 as drafted, 3 edited)?

Each pitch, the piece it references, and the verified contact route. Podcast-host drafts were redone on GPT, shorter, at your request.

Nothing has been sent yet.

Decision: Approved 8 as drafted, edited 3 before approving, rejected 14. 11 sent starting at 08:30, spaced through the morning rather than all at once.

Timeline

  1. · Email

    Operator replies by email with the plan and an estimate. Nothing will be sent without your approval.

  2. · Browser · Gemini Flash

    Research across recent coverage of personal AI agents and the named competitors, 83 pages visited, checking each candidate on their own site or byline page.

  3. · Browser · Gemini Flash

    25 candidates found, 22 verified with a name, outlet, specific piece and public contact route. 3 dropped, with reasons, for lack of a verifiable outlet or contact route.

  4. · Files

    Spreadsheet written: name, outlet, piece (linked), contact route, why relevant, verified (y/n).

  5. · Claude

    25 tailored pitches drafted, each referencing the specific piece, in your tone from memory.

  6. · Email

    Brief sent: 25 targets, 22 verified, 3 dropped (reasons attached), 25 drafts waiting.

  7. · GPT

    You ask it to redo the podcast-host drafts on GPT, shorter. It switches model for those 6 drafts, keeping the same context and tone; the other 19 stay on Claude.

  8. · Email · Held for approval

    Approval card sent for the reviewed batch: 8 approved as drafted, 3 edited then approved, 14 rejected. Nothing sent yet.

  9. · Email

    11 approved pitches sent, spaced through the morning.

Receipt

Steps
9 recorded
Checked
83 pages visited; 25 candidates found, 22 verified, 3 dropped with reasons.
Models used, and whose key
  • Research and verification (overnight): Gemini Flash (default key)
  • Initial 25 drafts: Claude (your Claude key)
  • Podcast-host drafts, redone shorter: GPT (default key)
Sent
Nothing sent overnight. 11 pitches sent from 08:30, spaced through the morning; the other 14 held back.
Held
14 drafts (not approved) and the 3 dropped candidates, none of them pitched.

Common questions

Do I have to keep the chat open for a task to finish?

No. Once you've sent a task, you can close the app, turn off your device, or start something else. Your operator keeps working in its own workspace and reports back when it's done.

How will I know when it's finished?

The result appears in the channel you asked from and in your task history. You can also ask your operator for progress at any point.

What if it needs my input partway through?

It holds at that step and tells you, then keeps working on any part of the task that doesn't depend on your answer.

Is there a limit to how long a task can run?

Most tasks finish in minutes to a few hours depending on the work. There's no fixed cutoff for a task that's making progress, and a task that stalls flags it to you and doesn't run indefinitely.

Hand off your first task tonight.

Tell us your email and what you'd hand off first. We'll send your access details and help you set up your operator.