For a job too big for one pass, your AI coworker plans a team of temporary helpers and runs them at once — 200 tickets grouped, 30 companies researched.
For a job too big for one pass, your coworker plans a team of temporary helpers and runs them all at once.
Two hundred tickets, grouped by theme. The phases are declared before any work starts — so you can stop it if that isn't what you meant.
Some jobs aren’t hard, they’re just large. Two hundred support tickets to group by theme. Thirty companies to research. Every page of a site to check.Working through those one at a time takes forever. Instead, your coworker writes a plan — the phases the job breaks into — then puts a team of temporary helpers on it, running side by side. Each helper takes a slice, does its part, and reports back. Your coworker pulls the results together into one answer.The helpers exist for that job only. They aren’t coworkers you hired, and they’re gone when it’s done.
Open your coworker, choose Capabilities, and look under Collaboration.
Dynamic workflows — Plan and run a team of temporary subagents for large jobs — parallel sweeps, pipelines, cross-checking. Powerful on big tasks and token-intensive.
In plainer words: it puts a team of temporary helpers on one large job, working side by side, and it costs more than an ordinary task.
It needs Working with code and files. Turn this on without that and it can’t do anything — the tab will tell you so directly:
Needs “Code and files” turned on above to do anything.
The warning appears on the row itself, so a switch never reads as on while nothing happens.
This is off when you hire a coworker, and it’s the more expensive of the two Collaboration switches. A run puts a whole team to work at once, so it costs considerably more than a normal task. Turn it on for coworkers that handle big jobs; leave it off elsewhere.
The run card is the good part, because it shows you the plan before any work starts.
1
The plan appears first
Your coworker declares the phases the job breaks into — before a single helper starts. You can see what it intends to do and stop it if that’s not what you wanted.
2
Helpers fill in live
A row per helper, each showing what it’s working on and its own progress. Rows complete as they finish, not in order.
3
Results come together
Your coworker takes what came back and produces the single answer you asked for.
A run that grows very long pauses and continues rather than failing, so a big job doesn’t get lost partway through.
Say how big the job is when you ask. “Go through all 200 tickets and group them by theme” tells your coworker to plan for scale. “Have a look at these tickets” doesn’t.
Cross-checking is an underrated use. Asking your coworker to verify one important conclusion three independent ways catches things a single pass misses.
A workflow that works well can be kept and run again later by name. See Slash commands.
How is this different from delegating to another coworker?
Delegation hands work to a real coworker you hired, with its own instructions and memory. This creates temporary helpers for one job and discards them afterwards.
How much more does it cost?
Meaningfully more — you’re paying for a team rather than one coworker. That’s why it’s off by default and why it’s worth reserving for jobs where doing it one at a time genuinely isn’t practical.
Can I stop a run partway?
Yes. Stop the task and the helpers stop with it. What already finished stays in the conversation.
What if one helper fails?
The rest carry on. Your coworker tells you which part didn’t come back rather than quietly returning an incomplete answer.
Do I need this for everyday work?
No. Most tasks are better served by an ordinary one. Reach for this when you catch yourself thinking “this is going to take all afternoon”.
