Hand off the job. Get your attention back

A background run takes a well-defined task into its own session. You keep working while it reads, edits and tests, then pick up a draft to review.

Install ML.ai All use cases

New sessionAutomationsReview

Sessions

Fix 401 in auth.test.ts
Plan the session store move
Add dueDate to tasks

Add dueDate to tasksacme/tasks0m 12s

Run the full test suite, fix any failures you find, and report back.

Started a background run.

It has its own session. You can keep working here.

Background run · Fix failing testsSucceeded

Ran npm test2 failed

Fixed the date formatting informat.ts

Re-ran the suitepassing

Draft ready in your composer

format.ts+1−1

AllowAlwaysDeny

Describe a task or ask a question

Mode BuildModel AutoEffortlowmediumhigh

Background runs3

Fix failing testsSucceeded
Update the API docsAwaiting approval
Migrate the loggerSucceeded

format.ts+1 −1 · draft

12export function formatDate(date: Date) {

13- return date.toISOString().slice(0, 10)

13+ return date.toLocaleDateString('en-CA')

14}

Who uses to delegate work

Anyone with a job that would otherwise mean sitting and watching a progress indicator.

Developers with long jobs
A full test run, a migration or a long fix that should not hold up the conversation.
People running things in parallel
Start a second task, or ask a quick question, while a run keeps working.
Reviewers who want a draft first
Results land as a draft in your composer, so you decide what to keep before anything is sent.

Delegating only works if you can still trust the result

Handing a job off saves your attention. These four things keep it from costing you control.

Main sessionfree

Background run · Fix failing testsworking

Two sessions, one of them free

Waiting on a job is dead time

A long job running inline holds up the conversation. In the background it takes its own session and the one in front of you stays free.

The panel wasreloaded

Fix failing testsstill running

Still running after the panel reloaded

The run survives the panel

It lives in its own session, so closing, hiding or reloading the panel leaves it running. Open the panel again and you can see where it got to.

Background run · Fix failing tests

Wants to write 2 lines in format.ts

AllowAlwaysDeny

The same prompt, in the background

Not watching it does not lower the bar

A background run hits the same Allow, Always or Deny gate as any other run. If it needs a decision, it waits for you.

Draft in your composer

Two tests failed because dates were formatted in UTC, a day off in other time zones. Fixed in format.ts.

format.ts+1−1edit, send or dismiss

The result lands as a draft

The result arrives as a draft, not a surprise

When the run finishes, its output appears as a draft in your composer. Read it, edit it or send it on in your own time.

Watching a job run versus delegating it

You kick off a long task, then sit watching a progress indicator, because the alternative is losing track of whether it needs you.

Terminal

$npm test → 2 failed

$fix src/utils/format.ts

$npm test → 1 failedRe-run

$npm test → passing

Twenty minutes of watching output instead of working

The task runs in its own session while you keep working. It waits for you when it needs a decision, and its result comes back as a draft when it is done.

Run the test suite, fix any failures, report back

Started a background run

Fix failing tests ·draft ready

format.ts+1 −1

13- return date.toISOString().slice(0, 10)

13+ return date.toLocaleDateString('en-CA')

Which agent actually does this

general

ReadRun testsWrite diff

general · Read · Run tests · Write diff

Runs the task in its own session

The same agent as a foreground request, delegated to a background session that survives the panel closing.

explore

SearchReadWrite

explore · Search · Read · Write

Optional, for a read-only job

Delegate a question instead of a change. It searches and reports back without being able to edit anything.

How a delegated job goes from hand-off to draft

A background run moves through four stages, and you can step away at any of them without losing the work.

Explorer

test/suite/format.test.ts
src/utils/format.ts
src/utils/date.ts
package.json

ML.ai ML.ai

general

Sent to the background: it gets a session of its own

Run the full test suite, fix any failures you find, and report back.

Mode BuildModel AutoEffortlowmediumhigh

1Hand off the taskOpen the background drawer with /background, choose Delegate a job, pick a delegate and describe the job. A new session opens for that job alone.

Explorer

test/suite/format.test.ts
src/utils/format.ts
src/utils/date.ts
package.json

ML.ai ML.ai

general

Run the full test suite, fix any failures you find, and report back.

Started a background run

Running in its own sessionkeep working here

Ran npm test2 failed

Draft ready, 1 diff

Describe a task or ask a question

Mode BuildModel AutoEffortlowmediumhigh

2It runs in its own sessionThe run is independent of the current panel. Closing or reloading the panel does not stop the work in progress.

Explorer

test/suite/format.test.ts
src/utils/format.ts+1−1
src/utils/date.ts
package.json

ML.ai ML.ai

general

The run found 2 failing tests and fixed the date formatting behind them.

format.ts+1 −1 · draft

12export function formatDate(date: Date) {

13- return date.toISOString().slice(0, 10)

13+ return date.toLocaleDateString('en-CA')

14}

AllowAlwaysDeny

Describe a task or ask a question

Mode BuildModel AutoEffortlowmediumhigh

3Keep working, check in when you likeAsk questions or start something new in your main session. If the run needs a permission decision, it shows up as its own status instead of blocking silently.

Explorer

test/suite/format.test.ts
src/utils/format.ts+1−1
src/utils/date.ts
package.json

ML.ai ML.ai

general

The run found 2 failing tests and fixed the date formatting behind them.

format.ts+1 −1 · draft

12export function formatDate(date: Date) {

13- return date.toISOString().slice(0, 10)

13+ return date.toLocaleDateString('en-CA')

14}

AllowAlwaysDeny

Allowed · format.ts saved, suite re-run green

Describe a task or ask a question

Mode BuildModel AutoEffortlowmediumhigh

4Take the result when it landsA finished run’s output arrives as a draft you can pull into your conversation or follow up on, never auto-sent without you looking at it.

Explorer

test/suite/format.test.ts
src/utils/format.ts
src/utils/date.ts
package.json

ML.ai ML.ai

general

Sent to the background: it gets a session of its own

Run the full test suite, fix any failures you find, and report back.

Mode BuildModel AutoEffortlowmediumhigh

Common delegation problems and fixes

What usually makes handing work off feel risky, and how ML.ai closes each gap.

Panel reloaded mid-runWith ML.ai

Fix failing testsstill running

Main conversationfree

Progress so farkept

The run carries on

Reloads that throw the work away

When a run is tied to the panel, a reload or a closed window ends it, and whatever progress it made goes with it.

How ML.ai ML.ai handles it

The job runs in its own session, so the panel can close, hide or reload and the run keeps going until it finishes.

While the suite runsWith ML.ai

Ask a questionanswered now

Start a second taskstarts now

The long jobruns on its own

Two things at once

A long job blocks the conversation

A big test-fixing job running inline means you wait, with no way to ask a quick question or start something else.

How ML.ai ML.ai handles it

The run takes the job off the main session, so the conversation in front of you stays free.

The finished runWith ML.ai

format.tsa draft

Summarya draft

Your approvalstill asked

AllowAlwaysDeny

Output you cannot adjust first

Output posted straight into a chat is hard to trim or correct before it goes further, say into a pull request description.

How ML.ai ML.ai handles it

Output arrives as a draft in your composer, and any edit still waits behind Allow, Always or Deny.

Where this stops

The real edges of delegating a job, stated plainly instead of left for you to discover.

Run waiting on you

AllowAlwaysDeny

It still needs your approval

A background run hits the same Allow, Always or Deny gate. Waiting on approval is a real status you come back to.

Same rulesLooser rules

Same rules, separate session

Same permission rules and command safety classifier as the foreground. It is a separate session, not a separate set of rules.

Clear taskVague task

Best for a well-defined task

A task you could hand to general directly. An ambiguous one is better in the foreground, where you can redirect it right away.

Apply edits to 3 files?

One diff across the schema, the API route and the task card.

route.tsschema.prismaTaskCard.tsx+2 −1

20export async function POST(req) {

21- const { title } = await req.json()

21+ const { title, dueDate } = await req.json()

22+ return createTask({ title, dueDate })

AllowAlwaysDeny

Build

Describe a feature in a sentence and ML.ai writes it across every file it touches, held behind one real diff before anything saves.

See how it works

Root cause in session.ts

The store returns null before the session write finishes, so the middleware rejects a valid token.

session.ts+2 −1

40export async function getSession(id) {

41- const s = store.get(id)

41+ const s = await store.get(id)

42+ if (!s) throw new SessionNotFound(id)

AllowAlwaysDeny

Debug

Give ML.ai a failing test, an error message, or a stack trace and it reads the execution path, finds the actual cause, and proposes a fix as a diff you approve.

See how it works

FAQs

Does a background run skip permission prompts since nobody is watching it live?

No. It hits the same Allow, Always or Deny gate as any other run. If it needs a decision, it waits for you.

What happens if I close the editor panel while a run is going?

The run survives the panel closing, hiding or reloading, because it is its own session. Open the panel again to see where it got to.

Does the result apply itself when the run finishes?

No. A finished run’s output lands as a draft for you to pull in or review, never applied or sent without you seeing it first.

Which agent should run a delegated job?

You pick it when you start the run. general carries out changes end to end; explore and architect read and plan without being able to edit.

Can I run more than one job at once?

Yes, up to the number of jobs your plan runs in parallel (2 on Infinite, 6 on Pro, 20 on Max). Each delegated job gets its own session, and a run that is waiting on you does not hold one of those slots.

Try today, or talk to us about what is next

Install the editor agent on your own machine, or book a call to talk through your team’s workloads

Download ML.ai Book a call

Hand off the job. Get your attention back

Who uses to delegate work

Developers with long jobs

People running things in parallel

Reviewers who want a draft first

Delegating only works if you can still trust the result

Waiting on a job is dead time

The run survives the panel

Not watching it does not lower the bar

The result arrives as a draft, not a surprise

Watching a job run versus delegating it

By hand

With

Which agent actually does this

Runs the task in its own session

Optional, for a read-only job

How a delegated job goes from hand-off to draft

Common delegation problems and fixes

Reloads that throw the work away

A long job blocks the conversation

Output you cannot adjust first

Where this stops

It still needs your approval

Same rules, separate session

Best for a well-defined task

Related use cases

Build

Debug

FAQs

Try today, or talk to us about what is next