Building Teams of AI Employees

A single AI employee can only do so much. When you pile on tools, the employee gets confused about which one to reach for. When you ask it to play three different roles, the answers get muddled. When the work takes many steps, context from earlier steps starts slipping away.

A team solves this the same way people do at work. Every team has one team lead who makes the plan and hands out the work, and team members (AI employees) with their own roles who actually do it.

When do you need a team?

These are the signs to break things up.

Too many tools: One employee has a dozen tools enabled and keeps fumbling which one to pick.
Roles blur together: “Research, write, design, and review” stuffed into one employee. Some parts come out sharp; others, sloppy.
The work takes many steps: A long pipeline where carrying every detail end to end makes answers slow or fuzzy.

Team lead and team members: the basic shape

You hand work to the team lead, who divides it among roles such as researcher, writer, and reviewer, then brings their results together.

When you create a collaborative team, a team lead is created along with it automatically. It’s named “Team Lead”, and you don’t need to hire it. Existing teams all have a team lead too.
Each team member does its own job with the tools that fit it. A smaller toolset means sharper work.
The team lead and every team member keep their own memory, so nobody gets bogged down in what the others are doing.
In the employee list on web, the team lead shows a Team lead badge next to its name.

What the team lead does

Takes in work and splits it. When you post to team chat without saying who should do it and no one else is a clear match, or the request spans several team members’ jobs, the team lead picks it up and hands the pieces to team members.
Plans toward goals. Give the team a long-running goal, and the team lead works out what to do next each round and assigns the work. It remembers approaches that were tried and didn’t help, and looks for a different direction.
Gathers results and decides what’s next. It collects what team members finished, passes it on to you, and judges what to do after that.
Sets the team’s operating schedule. If something needs checking on a recurring basis, the team lead can set up a scheduled run for itself. A recurring job a team member has to do isn’t scheduled by the team lead on their behalf; it’s handed to that team member as work so they set it up themselves.

What the team lead doesn’t do

The team lead isn’t there to do the hands-on work. It has no execution tools like browser control, running code, creating files, or sending things out. It sticks to team conversation, managing tasks, looking things up, and web search. Real work always gets handed to team members.

That means a team with only a team lead can make a plan for a goal but has nobody to give the work to. In that case the team stops, reports what kind of employee it needs, and waits for you. Hire the employee it needs and add them to the team.

The team lead can’t be replaced

The team lead is a fixed seat on each team.

You can change the team lead’s name and avatar.
You can’t delete it, move it to another team, pause it, or change its mission or tier. Where the delete button would be, the team lead’s settings say “A team lead can’t be deleted on its own. Deleting the team deletes its team lead too.” You also can’t drag the team lead to another team or out of its team.
A task handed out to a team member can’t be reassigned to the team lead. The team lead splits the work; it doesn’t take on the hands-on pieces. The other way around, a top-level task that started from a goal can only be owned by the team lead.
The team lead isn’t switched off automatically for being idle or running out of SU. If it’s still left switched off anyway, the server turns the same team lead back on when it’s needed. If the team lead still stays unable to work, the goal pauses and you get a “check the team screen for why it stopped” notice.
To get rid of a team lead, delete the team. Deleting the team deletes its team lead along with its members.

Four ways to build a team

Option 1. Choose “Hire a team” from Hire

Press the + Hire button at the top of the Home screen, and a dialog asks what you want to add. Choose Hire a team to open the team hiring sheet.

Understand the business: pick what you do (industry, the work that eats your time, and so on) with buttons, then press Build my team. We read your answers and put together a team for that work. Building the team doesn’t use any SU.
Confirm: review the lineup and leave out anyone you don’t need yet. Confirm, and the team lead and team members are all hired in one go.

Once hiring finishes, a goal suggestion card may show up in team chat. Work routing rules, which decide who picks up what, start out empty; with no rules, the team picks an owner on its own. You can add rules yourself in team settings (see below).

On web, the team lead greets you on behalf of the team, and a team setup card guides the required setup. Newly created teammates prepare their own responsibilities and personality, so you do not need to repeat the business introduction for each one.

Option 2. Use a button the manager shows you

While you don’t have any employees yet, the manager may greet you with buttons like “I’m starting a new business” · “Automate my existing business” · “Automate my own work.” Pressing one opens the same team hiring sheet as Option 1. Choose Automate my own work and you can get a team built for personal work, like a job search. Say you’re job hunting, for example, and the team it builds has employees who find postings and put together your application materials.

Option 3. Just create an empty team

At the bottom of the + Hire dialog, Just create an empty team → opens a form for a team name (required) and a description (optional). If your account is part of an organization, you can also choose right there whether the team belongs to you personally or to an organization. Confirm, and a team with only a team lead is created and you land on that team’s screen.

Use this when you want to group employees you already have into one team. Add team members the way Option 4 describes.

Option 4. Move existing employees into the team

To bring in an employee that’s unassigned or on another team, drag its row onto that team in the employee table. See Dragging employees between teams below for details.

If an existing team needs a new employee, hire one with Hire one employee under + Hire, then drag that employee’s row onto the team.

A team built any of these ways works exactly the same, and you keep managing it through team chat, the team task list, and settings.

Deleting an empty team

You delete a team from Delete Team in the team’s Settings panel (open Settings & details from the ⋮ menu at the right of the team chat header; it opens beside the chat).

An empty team (no employees besides the team lead) can also be deleted straight from the employee table. Hover over the team row and a trash icon appears next to Team Chat →; clicking it opens a confirmation. The icon only shows for empty teams you have permission to manage.

When a team is deleted, its team lead, team chat history, pinned items, and team skill pool are all cleaned up along with it.

Deleting from the Settings screen permanently deletes every employee still on the team, too. It can’t be undone. To keep the employees and remove only the team, first drag them out of the team into the Independent group in the employee table so the team is empty, then delete it (see Dragging employees between teams below).

On narrow screens (mobile web), the sidebar list also shows a delete icon when you hover over the name of an empty team (one with no employees other than its team lead). That shortcut only works on empty teams; if the team actually has team members, the delete is blocked.

Note: In the past, a team was automatically deleted when its last employee was removed. That no longer happens. An empty team stays around until you choose to delete it yourself.

Dragging employees between teams

In the employee table, you can drag an employee’s row to change its team assignment on the spot (on a touch screen, grab the handle at the left of the row or the avatar). Either direction shows a confirmation dialog before the change actually takes effect. The team lead can’t be moved to another team or released as a standalone employee.

Moving a standalone employee onto a team

Say Researcher is a standalone employee, not yet on any team. Drag its row onto the Content Team’s name row and confirm the dialog, and Researcher immediately joins Content Team, ready to take work through team chat and the team task list. If the employee was already on a different team, it leaves that team and joins only the new one.

Releasing a team employee back to standalone

The reverse works too. Drag Writer’s row (currently on Content Team) onto the Independent group and confirm the dialog, and Writer goes back to being a standalone employee that isn’t on any team. Its conversation history and settings stay exactly as they were, and dragging it back onto the same team at any time undoes the change.

On mobile: In the iOS and Android apps, the same gesture (press and hold to drag) is supported in List view. Board view is for viewing status at a glance and doesn’t support this. See Viewing the team board on mobile below.

Team chat: a place to give orders, not just watch

Every team has a Team Chat screen. It isn’t just a window onto what your AI employees say to each other: you can post to team chat yourself and direct the whole team.

Because team chat is a room where several people and employees can be talking, it’s laid out as a single-column roster instead of the left/right speech bubbles a 1:1 chat uses. Whoever’s speaking, an avatar, name, and timestamp line up along the left. Your own messages aren’t set apart by position, just by an accent-colored name and a highlighted avatar ring.

At the top of team chat, a progress bar shows only the work that is actually running right now (it shares that spot with the goal-and-budget card, and summarizes up to 3 steps). You can see what’s in motion at a glance, without scrolling back up. When the work finishes, it is highlighted briefly and then disappears on its own, without waiting for a new message. Work that stopped because it needs a person is not shown in the progress bar. Check and answer it in the Requests for you panel instead. Web, iOS, and Android behave the same way.

Use @mentions to hand something to a specific employee.

[You]: @Researcher please dig into “EV battery market trends” this week.

↑

This single line wakes the Researcher and starts the work.

Type @ in the composer and team member names autocomplete. Writing a name automatically wakes that employee so it can start working.
You can mention several employees in one message to wake them in parallel.
If you speak without an @mention, only one employee receives it, not the whole team. (Inside a thread that already has a conversation going, it works differently. See “Continuing in a thread” below.) Whatever you write in the team’s main chat always stays in the main chat and is handled as a new request, even if some work is already in progress. It is not moved into that work. The team checks whether it matches a workflow the team saved or a work routing rule, and on a team with several team members, it may pick the right owner automatically based on what you said. If no owner can be found, or the request spans several team members’ jobs (such as “handle everything from the quote request to the purchase order”), the team lead takes it and splits up the work. If no employee is currently active, the message is saved but nobody wakes up.
Employees also call on each other and share updates in team chat. When actual work gets handed over, though, it’s split into tasks rather than passed along in conversation (see “The team task list” below).

You see who will actually receive it before you send. When you type in the team’s main chat, a line right above the composer shows who a message would go to if sent as-is. If an @name you typed has a typo, or names someone who can’t currently answer (paused, or still in work setup), that shows up too, so you’re less likely to send something and only learn later that nobody responded.

Adding to work that is already underway

To keep talking about work that is already in progress, open that work’s thread and write there. A message you write in the main chat is never moved anywhere, so opening the thread is the clearest way.

In the main chat, click the thread link under a message (the line that shows the reply count) to open that work’s thread. When a thread is attached to a piece of work, a single line just above the link shows its status, title, and owner.
On a work card in the team task list, click Reply in thread to open that work’s thread with the composer ready to type. The card also shows the thread’s reply count and how many replies you haven’t read.
Click the work badge at the top of a thread to open the team task list with that work’s card highlighted.
When an employee asks you a question on a card and you click Respond, it opens that work’s thread instead of using the main chat composer, and you answer there. The same goes when you attach a file with your answer.

When a message wakes a run, a progress card attaches below it showing what step is underway. While it’s running, the card summarizes the last tool used; once finished, it shows something like “3 steps · 2 min”; expand it to see the full list of tool calls. You can stop an in-progress run right from the card with [Stop].

There can be a short wait between sending a message and an employee actually starting on it. On web, a line like “Researcher · 1 requests received · Waiting to start” shows near the composer during that wait, along with the elapsed time, so you can tell the request is waiting its turn rather than stuck. Once the work starts, that line goes away and the working employee’s avatar shows instead. In the rare case a run can’t start, the line changes to “1 requests could not start”, and you can close it with the × button next to it.

Mention conversations fold into a thread below the original message. Expand it to see the replies and the progress in between, in order. If there are threads you haven’t opened yet, a collapsed “Unread threads” card shows up at the top of the chat. It looks just like the other progress cards, so it doesn’t get in the way. Expand it to see, line by line, who started each thread and about what, and click a line to jump straight there.

(Web) If a lot of unread threads have piled up when you open team chat, a briefing dialog appears so you don’t have to open every thread one by one: it rolls up what happened while you were away, work by work. Anything that needs your input shows up first. Threads and work items follow, grouped as Waiting or blocked, Wrapped up, In progress, and Other conversations, each with its status, result, and a short summary. On a wide screen with several items, a contents list on the left takes you straight to any item. Press Continue in thread to jump there, or press Mark all as read to mark everything you’re looking at as read in one go. Just closing the dialog doesn’t change what’s read. While the summary is being put together, a “Summarizing threads” banner stays pinned briefly. Unread counts are also split between messages addressed to you and traffic just between employees, with the latter shown as a lighter gray number.

Team chat: one employee replying to another's mention. The original mention is quoted above, and the reply body @mentions the sender back

You can attach and send files in team chat too, with or without any text alongside them.

When you send a spreadsheet file such as Excel or CSV, an employee may offer first to turn it into a managed table. You don’t have to ask. Approve it on the card, and the file’s contents go into a table you can keep managing on the Files & data screen. If the file looks like something you only need to read once, the employee won’t offer, and sheets with a header row but no data, or files too large to bring in at once, don’t get an offer.

Continuing in a thread

Replying to a team message continues that message’s thread. Say an employee posts a report and you reply “just fix this part”: instead of landing in the team’s main chat as a brand-new request, it becomes a follow-up inside that report’s thread.

On web, hover over a message to see the Reply and Quote in main chat buttons. Reply opens the thread with the quote filled in. If you want to post in the team’s main chat with just a quote, without going into the thread, use Quote in main chat.
On iOS, press and hold a message for the Reply · Quote in main chat · Open thread menu.
On Android, tapping a message replies, and pressing and holding lets you pick Quote in main chat.
You can start a thread on a message that has no replies yet. Replying to a message that’s already a reply opens its thread from the first post.

Inside a thread, a message without an @mention goes to the one owner of its unfinished task, when there is one. Mention @Name to ask a specific teammate. Without an unfinished task, active employees who have already participated receive it. Mention @Name to choose who should reply. Open the 👀 or ✓ marker under your message to see who received it.

On web, use a card’s Sources links to open the related task or original conversation. You can expand a waiting progress card to inspect the request, and use Tab to accept the faint suggested text in the composer before editing and sending it.

You can also ask the owner in the task’s thread to stop the work. Requests that continue open work are kept distinct from separate follow-up tasks and subtasks.

Answering questions and finding what the team made

Question cards: when an employee needs to ask you something mid-work, a question card appears in team chat, and you can answer right there by picking an option or typing a reply. If you don’t have permission to answer, the buttons show as blocked.
Rate this result: hover over an employee’s result message (web) or press and hold it (mobile) and choose Rate this result to leave a 👍/👎 with a short note. It’s visible to anyone who can see that team chat, and the employee sees it as reference on its next run. Only the person who left it can delete it.
Canvas and web app chips (web): chips next to the composer reopen canvases and web apps made in this team or posted to it. No need to scroll up hunting for the card after a long conversation. Things an employee made in a private conversation don’t appear here.
Conversation search (web): use the magnifying glass in the team chat header (or Ctrl+F, Cmd+F on Mac) to search within the team conversation.

Posting something from a private conversation to the team

A canvas or web app made in your private conversation with an employee doesn’t show up in the team on its own. To share it with the team, press Post to team. You’ll find the button on the screen where a canvas is open, and in the web app list. The detail screen you get by opening a file from the Files tab of the Files & data screen has the same button for posting a file to a team (web, iOS, and Android, as long as there’s at least one team you can see).

Press Post to team and a dialog opens to pick the destination team and an optional caption.
Once posted, a card for it appears in team chat, and it also shows up in the canvas and web app chips next to team chat.
What gets posted is a card pointing to the original, not a copy, so edits to the original show up in the team too.

Only someone who manages that employee (or, for a file, someone who can open that file) can post, and only to teams you can see that belong to the same organization (or personal space) as the item. The default team has no shared conversation, so you can’t post there, and a file restricted to one team can only be posted to that team. Pressing it again for the same item doesn’t post it twice.

Seeing where you posted, and taking a post back

Press Post to team again on something you’ve already posted, and a Posted to list appears at the top of the dialog. Each team shows who posted it and when (or when it was withdrawn).

View in team: jumps straight to the post in that team’s chat. Right after you post, this link also appears next to a “Posted to [team].” message.
Withdraw: removes the post from that team only. It drops out of the team’s canvas and web app chips, but the original and its public link are not deleted and stay as they are. You’ll be asked to confirm first.
Post again: re-posts to a team you withdrew from. A new post message goes up in team chat.
Teams you haven’t posted to yet can still be picked from the team selector below the list.

Only the person who posted it, the owner of the original, or someone who manages the team (the team owner or the organization owner) can withdraw a post. If you don’t have permission, the Withdraw button doesn’t appear.

Mentioning team work in a one-on-one conversation

If you’re chatting one-on-one with a team employee and bring up work the team did, something like “How did that competitor research the team was doing turn out?”, the employee finds that work and shows a guide card in the conversation. The card shows the team and work name, its status, and who owns it. If several pieces of work could match, the employee asks which one you mean first.

Work still in progress (Received, In progress, Blocked, Needs your input): the Continue in thread button takes you to the team thread where the work is happening. Leave follow-up requests and answers there; the employee won’t carry on that work in the one-on-one conversation.
Finished or canceled work: View result (or View thread) takes you to where it ended, and Send a follow-up to the team opens the team chat composer with a “Follow-up to” marker attached. Write your request and send it, and it starts as new work linked to the original. The employee who picks it up builds on the earlier result instead of starting over.

The card is only a way to get there. The employee doesn’t post to the team on your behalf; you decide what to send from the team screen. The card only appears in conversations where you’re the one talking, and not for people who can’t see that team.

When the manager drafts a message for another employee

While chatting with the manager (the AI manager that guides you through the service and your employees’ status), if something comes up that another employee should handle, the manager can pre-fill that employee’s composer with a draft message and take you straight to that conversation. The manager never sends it for you; you’re always the one who actually sends it. If you already had a draft sitting in that composer, the new text is appended after it. The same goes for approvals another employee needs: the manager doesn’t click those on your behalf, it just points you to where the approval button is.

Seeing who picked up what

When a new instruction gets assigned, team chat shows a one-line note like ”→ assigned to @Name” right alongside it, so it’s clear at a glance who’s handling it without having to ask.

If the assigned employee needs a human call while it’s working (say, a question only you can answer comes up), the item shows a needs your input status in the team task list below, with a button you can use to reply. Anyone who can post in team chat can answer it. If more than one person answers at the same time, only the first reply that lands actually takes effect, and everyone else sees it’s already resolved. Once an answer comes in, only that piece of work picks back up; everything else keeps going untouched.

A started task opens its thread beside the chat

When a request you sent in team chat starts as a task (it becomes in progress, needs your input, or gets blocked), that task’s thread opens by itself in the side panel next to the conversation. Progress builds up inside the thread, so you can see the work moving without going to look for it.

On web, and only when there is room for a side panel. If a thread panel is already open it won’t open another, and once you close the panel it won’t reopen because of requests that were still waiting. The iOS and Android apps don’t open it automatically.
Small talk or a quick thanks doesn’t open the panel, since it isn’t confirmed as a task yet. Tasks that are already finished or canceled don’t open it either.
In team chat, the task line shows a Task label after the status chip, so it’s clear the title is the task name. If an employee is working in that thread right now, the line also shows a one-line note of what they’re doing. When more employees are working at once, a count like “+1” is added.

Picking a saved workflow instead of repeating yourself

Once a team has learned the order of a repeated job and locked it in, you don’t have to say the same thing again. Just pick the saved workflow from a list and start it. How to run one is covered right below, in “The team task list.”

Setting work routing rules

Under Work routing rules in team settings, you can write rules like “messages with these keywords go to this employee.” When you post without an @mention and a keyword matches, that rule’s owner takes the work, and rules higher up win. With no rules, the team picks an owner on its own.

Team settings still has the old Coordinator field, but because the team lead now handles planning, assignment, and gathering results, it only shows a note that a separate coordinator can’t be set.

Jumping between employee chat and team chat (web)

On an individual conversation with an employee that belongs to a collaborative team, a chip with the team’s name sits next to the employee’s name. Click it to jump straight into that team’s team chat. If team chat has any unread messages, the chip shows a small badge.

The team chat screen has a back button that returns you to wherever you came from.

If you got there by clicking the team chip from an employee’s chat, back takes you to that same conversation.
If you got there some other way (Home, a shared link), back takes you to Home.

This is web only for now; it isn’t in the iOS or Android apps yet.

The team task list: seeing what the team is doing

Hand a team something to do, and that work becomes a task that stacks up in a hierarchy. The first thing you asked for is the top-level task; when its owner (usually the team lead) splits it up, team members’ subtasks hang underneath. The team task list lets you see and manage that structure at a glance.

On web, a chip next to the composer opens it as a side panel without leaving the chat screen. Team scheduled runs have their own chip too.
On iOS and Android, tasks, roster (graph), and scheduled runs are combined into one screen called the Team task list. A chip above team chat opens it, and you switch between the three with tabs at the top.

What the board shows for each item:

Status: whether it’s waiting, in progress, blocked, needs confirmation, or done.
To confirm: when a task is waiting on you, a box appears at the top of the card with who is asking, what they need, and the buttons that finish it. Its chip tells you the kind. A Question is something the owner is asking, so Respond or press Skip and continue. Paused means this request used up the times it can hand work to teammates; Continue picks it up as a new request. Check result means a result such as a quote has no confirmed values, so it wasn’t closed as done, and Notice covers other reasons, like the owner being turned off. When a subtask is asking, its question moves up, and the team lead’s line that repeats it folds into Lead’s note. Answering that question releases the lead’s confirmation too.
How much it’s used: one line showing how many subtasks exist, SU spent so far (including subtasks), and time left before the deadline. A value turns to a warning color as you approach its limit.
Owner: an avatar and name showing who currently owns the work.
Number: a top-level task shows a number in front of its title, assigned in order within the team. Without a prefix it looks like #31; with one it looks like RFQ-31 (see Task numbers and prefixes below). Subtasks have no number.
Mail source: a task that started from an incoming email shows a line under the title with sender · email subject, so you can tell which email each task belongs to even when several tasks share the same name.

Two actions are available from the ⋯ menu at the top right of each card. A task that has no owner and needs a new one shows Reassign as a button in its To confirm box.

Reassign: hand it to a different team member. A piece of work a team member received can’t be handed to the team lead.
Stop: cancel a task that’s in progress. Any subtasks get cancelled along with it, and work already underway is interrupted (this can’t be undone, so a confirmation dialog appears first).

Finished top-level tasks collapse automatically to keep the board tidy. A finished top-level task that was split among teammates has a Save as team workflow button under it from the very first completion (web and the latest iOS and Android apps; in older app versions it only appears after the same shape of work has repeated). When the same shape of work comes up again, a hint shows how many times that task type has come in recently and the button is highlighted, but you can press it before that. Unlike a personal employee’s workflow, there’s no composer draft: it saves the moment you press it, with a “Saved this order as a team workflow.” notice and a View saved workflows button. The board and saved-workflows list refresh right away so you can run it from the same spot immediately. A task that can’t be saved has no button. If it’s already saved as a team workflow or the workflow is locked, that is shown in the button’s place. If nothing was split among teammates (work one employee handled without splitting it into sub-tasks), or no step has an assignee or was finished all the way, the reason is shown only after the same shape of work has repeated enough. For the whole flow compared with a personal employee, see Save your first task as a workflow and reuse it.

On web, a result card may offer to Save this way as a workflow even when one teammate completed the task alone. This saves that employee’s regular workflow, rather than a workflow split among teammates. Writing up and saving the workflow may use SU; saving fails if there is not enough SU. Where relevant, you can also review a proposal to save the result to a table.

Task numbers and prefixes

A top-level task gets the next number in its team automatically when it’s created. You can’t set or change a number, and once assigned it stays. Top-level tasks the team already had were numbered too.

With a prefix the number reads like RFQ-31; without one it reads #31. A prefix is 1 to 3 letters or digits starting with a letter, and lowercase is saved as uppercase. Saving it empty removes the prefix. When you change the prefix, numbers already assigned show with the new one.
Where to set it: the person who manages the team does it on web in team settings under Task number prefix (Edit button), and on the iOS and Android apps in the field of the same name in team chat settings. A prefix in the wrong format isn’t saved and you’ll see a message.
Calling a task by number: in team chat, say something like “What’s the status of RFQ-31?” (#31 works too) and the team’s employee looks the task up by number and picks it up from there. The task list screen doesn’t have a separate number search box.

Choosing how a team member delivers results

Not every team member finishes work by making a new file. One reviews what the person before them produced; another’s job is to send emails. So for each team member you can set how they finish the work they get from the team lead.

Option	For members who…	Counts as done when…
Make a deliverable (default)	create new results like reports or tables	there is a newly made file, canvas or link
Review	check the earlier step’s deliverable	they checked the received file and passed it. If something needs fixing, it goes straight back to the earlier step’s owner to redo
Take action	do something outside, such as sending email	there is a record of the real action, like a sent email
Report in text	report back in writing, without a file	they delivered a written result

Where to change it: on web, pick it under How results are delivered on that employee’s Settings screen and save it with the other settings using the save bar. On the iOS and Android apps, pick it in the item of the same name on the employee settings screen.
Who sees it: only regular team members of a team that isn’t the default team. It doesn’t appear in the settings of the team lead or of standalone employees who aren’t on a team.
Review can only approve a deliverable that came from the earlier step. It can’t approve and pass along a file like the original received from a client.
Take action isn’t finished by text that just says “sent”. There must be a real send record; if there isn’t, the team lead reads the written result and decides. For example, if a person said “don’t send the email this time”, the lead takes that instruction into account too.
When the team lead splits up the work, it can also set a different option for each part. The option you pick in employee settings is the default when no part-specific one is given.

When the top-level task finishes: a “Next steps” suggestion

When the very first thing you asked for (the top-level task) finishes, team chat sometimes shows a Next steps card: the team’s own suggestions for what to do next, based on this result.

Check the suggestions you want and press Do selected work, and a new task is created for them right away. If a suggestion is really a long-running goal, it instead takes you to the Set up objective screen to review and confirm the details before it starts.
If you don’t need any of it, press No need, or Pass for now if you’d rather decide later.
This decision is shared across the team. If someone else who shares the chat already answered, opening the card later just shows you the outcome.

Running a saved workflow directly

If the team has any workflows locked in, a Saved workflows list appears at the top of the board. Check the name and owner, then press Run to start it the same way, without saying it again. If a step inside the workflow can’t start right now (say, its assignee isn’t currently active), the Run button is disabled and the reason is shown alongside it.

On mobile: iOS and Android also let you pick a saved workflow from the list and run it, and work started this way carries a “Started from a saved workflow” origin tag so it’s clear it wasn’t a fresh human instruction. Saving a finished top-level task as a new workflow with Save as team workflow works the same way as on web, from the first finished task, in the latest iOS and Android apps. For where to save and what each option needs, see “Pick by what you want to do” in Save your first task as a workflow and reuse it.

On web, task cards show the workflow used and its starting owner. If a saved workflow has several possible tables to pass to the next step, someone who manages the team chooses one first. Export to work portal requires the organization owner, an organization space, and an open work portal. The screen lists any missing requirements or workflow items to fix.

Sending files between AI employees

An employee can hand off a file it created (a document, an image, code, whatever) directly to another employee. For example, a researcher can pass a finished report to the writer who picks up where it left off.

The sending employee attaches the file to a team chat message and @mentions exactly one employee to start the handoff. You can’t send the same file to several people at once; send it to each one separately if you need to.
The file doesn’t move the instant the message is sent. It’s copied onto the recipient’s own AI work computer the next time that employee starts a run. The team chat shows the filename and size right away, so you can see what was sent even before it actually arrives.
A single handoff can be up to 500MB, and delivered files are automatically cleaned up after 7 days.
Folders holding an employee’s internal settings or credentials are automatically excluded from what can be handed off, so they can never end up with another employee by accident.

A skill install approved in the team conversation, shared across the whole team

Getting an AI employee to install a new skill usually needs a human’s approval. When someone with approval permission answers an install card in the team conversation, that approval is recorded at the team level. After that, the team won’t ask again when a different team member wants to install the same skill. The same goes for a retried run after a failure.

What’s shared is only the decision that this particular skill is okay to install, not the whole roster’s install lists. Each team member still installs the skill into their own list only when their own work needs it, and it never shows up for a team member who has no use for it.
It shows up as a card in the shared team conversation that multiple team members can see, and the card records who approved or denied it.
On an organization-owned team, only an organization owner can respond to this kind of team-level approval. On a personal team, only the team’s owner can respond.
An install card in a one-on-one chat with an employee is also recorded for the whole team when the organization owner (or, on a personal team, the team’s owner) approves it. If a card asking only to install that skill was waiting in the team conversation, it’s cleared too. An install that any other member approves in a one-on-one chat applies to that employee only. Even if a teammate already uses the skill, another employee installing it is asked again unless there’s a team-level approval record.
If one card bundles several skill installs, only the skill shown on the card is recorded at the team level. The other bundled skills aren’t recorded for the team, so another employee installing them is asked again. When a skill needs a team record to install, such as one whose version changed, you’re asked again even in this run.
Employees on the default team don’t get team-level records. The default team is where employees who aren’t grouped into a team live.
A card that’s already sitting open on another screen resolves in real time the moment someone answers it, and you’ll see the same outcome whether you refresh or come back later.

Shared team notes: notes the whole team uses

Employees use two kinds of memory.

Memory type	Who can see it	What it’s for
Personal memory	Only that employee	Its own style, know-how, scratch notes
Shared team notes	The team lead and every team member	Facts, decisions, and standards everyone needs

Employees write down what the whole team needs to know in the shared team notes on their own, and refer to each other’s notes on later work. One employee’s personal memory can’t be opened by another, so anything the team needs to know together goes in these notes.
To see what the team is working on, use the team task list, not shared team notes; for goals, use the goal panel.

Viewing and editing the shared team notes yourself

You can also open shared team notes and edit them yourself. They’re a good place to write down standards employees must follow, like “Always put reports in a table” or “Discounts go up to 10% at most.”

Open the team settings. On web, choose Settings & details from the ⋮ menu at the top right of the team chat header (it opens as a panel beside the chat); on iOS and Android, tap the team name at the top and choose Settings from the menu.
Under Shared team notes, press Open (on Android, Open shared team notes; on iOS, tap the row itself) to see the list of notes.
Use Add item at the top of the list to create a note, and each note’s Edit and Delete buttons to change or remove it (on iOS, swipe a note to the left to reveal both buttons). Each note has a name, an optional description, and content.

Each note shows whether it was last written by a person or an employee.
Employees can’t delete a note a person wrote. An employee can still rewrite its content (which switches the label to “employee”), but a delete is refused. Employees also know a person wrote that note when they work with it.
People can edit or delete any note, no matter who wrote it.
If someone or an employee changes the same note while you have it open, saving shows “Someone or an employee changed this item in the meantime” and nothing gets overwritten. Press Load latest to review the change, then save again.
Anyone who can see the team (the team owner, or organization members for an organization-owned team) can edit its shared team notes. The default team has no shared team notes, so this section doesn’t appear there.

Giving a team a long-running goal

Give a team a goal such as “grow subscribers to 100k,” “make a competitor comparison report,” or “keep the order page responding quickly.” The team lead plans the next steps and assigns work to team members. When the approach is unclear, it first identifies what needs to be found out.

Turn your goal sentence into success criteria

In the goal editor, describe the result you want and press Draft criteria from the sentence. For example, “make a report comparing 10 competitors by Friday” lets you review and edit the deliverable, completion criteria, and deadline read from your sentence. Suggested values need your review; Not set marks something still missing. Check the type and date before saving too.

Objective type	What to review
Reach a number	How to measure, whether the number should rise or fall, the target, and its unit. You can also track without a target or rate reports yourself.
Finish a deliverable	What to produce, where to deliver it, completion criteria, and how to check them. No unnecessary target number is required.
Keep a state	The condition, allowed deviation, check interval, operating period, recovery condition, and permitted response. One normal reading does not mean the goal is achieved.

If the criteria are unclear, choose to save only the intent and let the team lead prepare it. Criteria you have already chosen are saved while preparation looks into missing materials or ways to check progress. A repeating time such as “every day at 7” is not a goal deadline; set it up as a separate scheduled run.

Starting from a goal suggestion card

After you hire a team, a “Set what this team works toward?” card may arrive in team chat. It’s the team suggesting goal options based on what you answered while hiring. The old step where you typed a goal sentence in the hiring sheet is gone; goals are now set from this card.

Review the option’s success criteria and deadline, and how it will be measured for a numerical goal. Some options ask you to rate each report with 👍/👎.
For an option ready to start, check I agree to autonomous operation and press Start now. The box is unchecked by default; saving without checking it does not start operation.
Options needing preparation lead to Let the team lead prepare it or Save the criteria. These save the goal and criteria or authorize preparation; they do not sign off on autonomous operation.
Press Later and nothing is created; Show the suggestions again brings the card back any time.
Each card can only be started once. If someone else already started it, you’ll see “That suggestion was already accepted.”

Teams without a goal yet

A team with no goals keeps a line reading “This team hasn’t decided what to work toward yet” on screen. On web it sits under the goal and budget card at the top of team chat; in the iOS and Android apps it’s in the goal panel. It stays until the team has a goal.

Enter the address of a web page that shows a number, or one sentence about what you want to grow, and send it. The team posts a matching goal suggestion card in team chat. With an address, it actually opens that page to find the number, so it takes a few seconds.
If no number is found at that address, you’ll see why and a prompt to try again.
A suggestion card can arrive shortly after hiring finishes. If no card can be created, the team lead goes looking on the web for a way to measure progress and posts a card if it finds one. If it can’t tell which page belongs to your business, it may ask you. If it finds nothing, it finishes quietly and the line above stays.
When the goal panel is empty, the Get goal suggestions button gets you fresh suggestions.

The people who can request suggestions or start a goal from the card are the same people who can change a goal, listed below. For a goal that’s already set, the Automatic operation switch in the goal panel turns autonomous operation on or off; with it off, the goal stays in place and the team only does requested work and scheduled runs.

Review preparation and start operation

Preparation checks the success criteria, materials, available team members and tools, required permissions, and ways to verify results. Agreeing to preparation is separate from signing off on operation. Before that sign-off, preparation is limited to reading and research; it does not authorize logins or external changes. Preparation also uses SU and requires the team’s automatic operation to be on.

Read the preparation card’s confirmed facts, small check, remaining unknowns, first task, and what is needed from you. Reading one number successfully does not mean all the work is ready. If materials, an account connection, or a new team member are needed, follow the card or goal overview’s guidance first.

The card distinguishes Ready, Check pending, and Waiting on you. Work outside preparation permissions, such as creating and reviewing a sample deliverable, remains a first task after operation starts.
Confirm saves the preparation result. When starting is available, Confirm and start also starts operation. Review the scope you are allowing before choosing.
A state goal’s first observation checks preparation. Whether the condition held over the agreed operating period is assessed separately.

How a goal runs

Starting a goal involves confirming, once, which accounts or channels it can touch and how far it can go (making videos, publishing them, running the channel, say). Inside that agreed scope, it doesn’t re-ask for approval on every individual action like a publish. Anything outside that scope (an account you didn’t authorize, a publish you didn’t approve) gets blocked automatically right before it happens and waits for a human to weigh in.
The team lead makes the plan and assigns work to team members. It uses research or small trials to investigate uncertainty, then uses the results and remaining unknowns to inform the next approach. Clear tasks do not each need a separate investigation or prototype.
Earlier attempts, reasons for rejecting an approach, and people’s answers remain available to inform later plans. Retries are linked to new evidence or changed conditions.
Changes to the goal, success criteria, deadline, or permitted scope are a person’s decision. The team lead adjusts the approach within the agreed scope.
If the team lead can’t work, there’s no team member to take the work, or planning keeps failing, the goal pauses and waits for a human. Fix the cause and pick it back up.
Only the team’s owner, the organization’s owner, or an operator explicitly designated for this can change or pause a goal. Pausing stops new tasks from being created; anything already running winds down at a safe stopping point.

Use evidence to decide and accept results

Open Show evidence and samples on a decision card to review linked research, files, or small samples, the effects of each choice, and the work that follows. Choose I have no material to decide if evidence is missing, or I won’t decide this if you decline. If research or comparison work must finish first, the decision waits for it.

You get notified the moment a decision card goes up. Decisions come in three grades:

Grade	48 hours with no answer	Answering later
Auto-advance	proceeds with the recommended option and closes	can no longer be answered
Provisional	proceeds with the recommended option for now	your later answer still reopens it for review
Person only	doesn’t proceed, keeps waiting	only moves forward once you answer

A decision that changes the goal itself is always person only, so it never proceeds without an answer. For a decision about the approach, the team lead assigns the grade, and if it leaves the grade blank, it defaults to provisional. Silence does not authorize a wider scope of action.

Before accepting results, open Show results and evidence to inspect the final work, judgments for each completion criterion, and remaining limits. You can expand long result text or open linked files, canvases, and links. The display distinguishes a check that a file exists from a judgment that its contents meet the criteria. Check the status if an original changed, was deleted, or is not accessible to you.

Review individual results using Accept or Send back when those actions are shown. Give a reason when sending a result back. A result already used by dependent work cannot be sent back. If a newer result arrives while you are reviewing, read it before answering.

For a deliverable goal, choose Confirm achieved or Request changes. When a state goal’s operating period ends, review its record and choose to confirm achievement, extend the period, or end it. If the goal changed, answer the latest confirmation card. If changed results prevent you from confirming an older goal check, choose Request changes so the team can provide a new goal check.

Check progress and cost in the overview

Open a goal’s overview to see Now, What we learned, Progress, Your move, and Next report. Preparing, waiting to start, operation being off, and being paused are distinct states. Depending on the state and your permissions, you can review preparation, start, pause, resume, or open decisions and results. Missing records or times are not replaced with invented completion percentages or estimates.

On web, tabs at the top of the goal screen switch between Overview, Path map, Task structure, and Decisions & evidence. The iOS app has two tabs, Overview and Decisions & evidence; the Android app has no tabs and instead shows the overview and decisions and evidence together on one scrolling screen. Neither app has the path map or task structure view yet.

Cost and limits shows total SU used and the preparation cost included in that total. The current path reference budget is a planning guide. The new-run stop threshold prevents new planning once spending reaches it; finishing work already in progress can take spending over it. You can also check the automatic run AI usage limit (last 24 hours) that applies to the team lead, and its usage. For a managed service organization, the organization’s limit is the one that applies.

Seeing how the team is doing at a glance

In the employee table, a team appears as a single group. The team’s name row shows the team’s totals for the last 7 days (runs, success rate, approval requests and expirations, SU, and so on) under each number column on a wide screen, or in one line next to the team name on a narrow screen, with one row per AI employee underneath. Each employee’s Status cell shows badges like awaiting approval, run failed, or paused, plus warning signals like failures in a row or out of memory, so it’s obvious where attention is needed. Next run, 7d runs, Success, Approvals · expired, 7d SU, and Note show from the start; Affiliation, AI performance tier, Skills · Tools, and Done are columns you can turn on with the [+] button at the end of the table’s header row (see Managing Employees on the Home Screen).

Team board: the team name and headcount with a summary of work done this week, checks requested, self-started share, and 7-day SU along the top, and cards for the three employees laid out underneath

Searching the board

Typing in the search box on the employee table’s top row searches more than just employee names. It covers conversation, task, and canvas content too.

Matches on an employee’s name, mission, description, or team name narrow the table’s employee rows to fit the search term.
Below that, a list of conversations, tasks, and canvases that mention the search term follows, showing which employee and which message or task it came from.
Clicking a result jumps straight to where it actually happened: the team chat, that employee’s own conversation, or the task itself.

You need at least two characters before a search starts.

Viewing the team board on mobile

The iOS and Android apps also let you switch between List and Board views near the top of the Employees tab. Board view shows per-team summary cards, and creating a new team or deleting an empty one both work from either view.

Dragging an employee to move it between teams or release it (see Dragging employees between teams above) is only supported in List view. Press and hold an employee card, then drop it on a team name to move it to that team, or on the Standalone employees area to release it. Board view is for viewing status at a glance and doesn’t support this gesture.

Transferring a personal team to an organization isn’t available on mobile yet; that’s still desktop web only.

What you see on the team page

The team’s default screen is Team Chat. There’s no tab row at the top of team chat, and how you reach the rest differs between web and the mobile apps.

On web, chips next to the composer open a side panel without leaving the chat screen.

Chip	What’s there
Team task list	The team’s task structure (top-level tasks and their subtasks).
Team scheduled runs	Every scheduled run across all employees on the team, in one place.

Settings & details in the ⋮ menu at the top right opens the settings panel beside the team chat (you stay in the conversation): team name, monthly SU limit, work routing rules, shared team notes, team deletion, and other admin options. Scheduled runs also opens as a panel in the same spot, from the chip above the composer.

Graph opens from the Open screen list on the path bar above the team chat, or from the search window, and appears as a side panel without leaving the conversation. People, the team lead, and team members show up as dots, with arrows drawn from whoever actually created a piece of work to whoever took it on. A thicker arrow means more work has been handed that way, work in progress shows as a flowing dashed line. Clicking a dot or an arrow shows a summary of that employee or that relationship below. This screen is read-only, so you can’t add or remove exceptions here. Work-handoff exceptions are managed from the Team members screen in the header’s ⋮ menu.

On iOS and Android, instead of separate chips, a single chip above the composer opens the Team task list screen, with tabs at the top for Tasks · Roster (graph) · Scheduled runs (see “The team task list” above for details). There’s no gear icon for settings either. Instead, tapping the team name at the top opens a menu right there, with links to the graph, goal & budget, recent activity, scheduled runs, and settings screens. Goal & budget opens right in place; the others take you to their own screen.

The roster shows each employee’s recent activity (a status icon, last response time), so it’s easy to see at a glance who’s working and who’s idle.

Teams on an organization account

If your account shares an organization (a shared space for a company or department), the employee table splits into Organization and Personal sections. Teams you create live under the Personal section by default, with team names appearing directly under each section. An “Independent” group label marks employees that aren’t on a team, and only appears when teams and standalone employees are both present.

In the employee table, you can grab a personal team’s name row and drag it onto the Organization section to transfer that team, and every employee on it (team lead included), to the organization. Instead of dragging, you can right-click the team’s name in the inbox Employees & teams list, or open the ⋮ menu at the top right of the team chat, and pick the organization under “Transfer to organization” (the default team and teams already in an organization don’t have this item). If even one member can’t be moved, nothing is transferred, and you’re told which employee blocked it and why. Before the transfer completes, a confirmation dialog spells out what’s about to happen:

This cannot be undone. A team transferred to an organization cannot be moved back to being personal.
Everyone in the organization will use the team and its employees, but only team conversations and employee history from after the transfer are visible to them. Only you, who transferred it, can still see the earlier history.
The archived conversation files and attachment copies the employees referenced are deleted during the transfer. You need to review every team member’s memory list, and any pending task or approval must be finished first.
You can choose which files move to the organization from those posted to the team or made by team members. Files you don’t pick stay in your personal space and won’t open in the organization.
“Next steps” proposals the team posted before the transfer that haven’t been answered yet are closed during the transfer. Ask the team again afterward if you still need them.
Ownership moves to the organization, so it stops being your personal asset.

This is currently only available in the employee table on desktop web, not yet on mobile.

If your account isn’t part of an organization, this feature doesn’t appear, and teams work exactly as before.

Team chat on an organization-owned team is shared by everyone in the organization. As described above, another person’s messages show that person’s name and avatar, so even with several people posting to the same team chat, it’s always clear who said what.

Standalone employees can’t use team features

An employee in the Standalone employees group on the employee list doesn’t belong to any collaborative team yet. That means it can’t use the following features.

Team chat (read/write), the shared team notes, the team task list, and team goals are all unavailable.

If you need a collaborative team, create a new team and move your employees there.

How are costs handled?

Every task inside the team is paid for from the team owner’s SU (yours), including what the team lead spends on planning and conversation. AI employees doing a lot of work can burn SU quickly, so keep the number of employees and their enabled tools to what you actually need.

Your headcount only grows when you hire or install a template, or when a new team brings its team lead along. Team employees can’t add colleagues on their own. Hiring a single employee brings up one confirmation, “Hire a new employee?”, and nothing is created until you press Confirm. That gate exists because each employee is always on and keeps consuming SU.

Every plan lets you own and run employees, including team leads, without a headcount limit. Repeated requests to create teams or hire employees in a short time can trigger a request rate limit, requiring a short wait.

Advanced

The section below isn’t needed for normal use. Read it only when a team is behaving unexpectedly, for example when a message isn’t reaching the employee you meant.

The team lead’s tools and tier

The team lead’s tools come from a fixed allow list: team conversation, the shared team notes, the team member list, splitting / reading / resuming / closing tasks, looking up the file list, data, and canvases, searching conversation history and memory, web search, managing its own scheduled runs, proposing a table, and asking you questions or sending cards. Proposing a table is only for suggesting that data someone brought into team chat be kept as a table; filling the table is split out to teammates. A tool switched off by employee settings or tier stays off even if it’s on this list. Execution tools like the browser, running code, writing files, sending things out, and external tools (MCP) aren’t on the list.
A scheduled run the team lead sets up runs only for the team lead itself. Putting a team member’s recurring job on the team lead would leave it with a lead that has no execution tools, so it’s handed to the team member as work and that team member sets it up.
The team lead’s base tier is fixed at Standard and can’t be changed. The turns where it plans toward a goal or works through goal preparation run at the Ultra tier, so those turns can use more SU than ordinary conversation. When the team lead or a team member starts designing a table or changing its structure, the rest of that turn also runs on Ultra regardless of their setting.
The team lead doesn’t do self check-ins (waking up on its own to look for work). It only moves on goal rounds, messages it receives in team chat, and its own scheduled runs.
The team lead isn’t switched off automatically for inactivity or running out of SU. If it’s still left off for one of those reasons, the same team lead is turned back on when a plan is needed. A team lead that an administrator switched off stays off, and since no other team member takes over planning in the meantime, the goal pauses.

Task statuses and limits

A task is always in one of these states: waiting, in progress, blocked, needs confirmation, or done (or cancelled). Whether something counts as “blocked” or “needs confirmation” isn’t decided by the server. The employee doing the work decides, because only it can tell from the conversation whether a peer could unblock it or only you can.

There are limits in place to prevent runaway spending: up to 5 direct subtasks under a single task, a cap of 5 US dollars in underlying cost on the whole task structure (about 200SU at the current rate, subtask runs included), a 30-minute deadline per subtask, and a 24-hour deadline for the top-level task as a whole. A top-level task that still has an unanswered confirmation request in its structure doesn’t expire right at 24 hours and waits for a human instead, but that wait is capped at 72 hours. These are fixed limits to prevent runaway behavior, not a plan feature, so they’re the same on every plan.

What employees say to each other

This section covers messages between AI employees in team chat. When you, a human, @mention someone in team chat, that employee always wakes up regardless of the limits below.

An employee’s post to team chat is one of two kinds. Share only leaves the message and wakes nobody; the recipient reads it on its next run. Ask wakes the named team member in that thread to get an answer.
An ask doesn’t create a task. Real work is always handed over by splitting tasks, and an owner who hasn’t split its own work yet is refused if it tries to push the work onto a peer with an ask.
A single thread can hold at most 6 asks.
One message can wake at most 3 employees, a chain of employees calling on each other runs at most 3 hops, and the same (sender, receiver) pair has a 60-second gap.
The old rule that “a connection must exist to wake someone” is no longer used. Instead, only pairs you put on the team’s block list can’t hand work to each other.

How an unmentioned message picks its recipient

A message without a mention in the team’s main chat goes to one recipient, decided in this order:

If what you said is the same job as a team member’s saved workflow, that workflow’s owner.
If there are work routing rules and an active owner’s keyword matches, that owner (rules higher up win).
If there are active team members besides the team lead, the team picks one automatically based on what you said and on the work each member handled recently. If no one fits, or the request spans several team members’ jobs, it moves on to the next step. A request for a single step still goes to that step’s owner, even when it mentions an earlier or later step.
If that still doesn’t settle it, the team lead (when active).
If the team lead isn’t active, whichever active team member was created first.

If no employee is currently active, the message is saved but nobody wakes up. This isn’t stored anywhere; it’s recalculated fresh each time based on who’s active right now.

The order above only applies to messages in the team’s main chat. A message without a mention inside a thread goes only to the owner of the unfinished task that started in that thread, if there is one. The owner knows how the handed-out work is going and asks other team members directly when needed. To hear from a specific team member, call them with @Name. In a thread with no unfinished task, the message goes to every active team member who has spoken in that thread at least once. If nobody has spoken yet, mention @Name to choose who should reply. Either way it doesn’t create a work assignment (which is why it gets a 👀/✓ marker instead of an ”→ assigned to @Name” line). In the iOS and Android apps, the message box shows who will receive it before you send: “Message @Name…” or “Message this thread…” in a thread, and “Message the whole team…” in the team’s main chat (on the web, this shows above the message box while you type). If only one member receives it, that one is guided to reply as usual. If the thread’s participants can’t be looked up, the message goes to one recipient using the order above.

Work routing rules start out empty, including on a team made with the team hiring sheet. You can add, remove, and reorder them yourself in team settings.

Because this kind of message comes directly from a person, it always runs. It isn’t subject to the budget or automatic-run limits that apply to messages between employees. Child runs started while handling it, however, can be held by the automatic run limit. Progress cards in the team chat run history eventually show “record expired” after enough time passes, but the final reply message itself stays in the chat regardless.

The waiting-to-start indicator

On web, the waiting-to-start indicator comes back after a page refresh or a reconnect. An indicator left behind by a run that never started is cleaned up automatically after 24 hours. This indicator is web only for now.

How the goal suggestion card works

Goal suggestions are prepared while the team is being installed. If preparation takes longer, the card can arrive after installation finishes. Deliverable and state goals can offer success criteria even without a way to measure a number. If no suggestion card can be created, the team lead gets a run to look for a way to measure progress. That run also uses SU and is spaced out so it does not repeat too often.
Whether what you typed is an address or a sentence is decided by the server. With an address, that address is the only candidate, and a card is created only if a number was actually read from the page. That value becomes the card’s current value and the starting point.
Starting a goal from the card uses what the server recorded when the card was first created. You can’t change the contents on screen while the card is open and start it that way.
If a table with the same name already exists, it isn’t overwritten; the existing table is used as is.
By default, goals measured by a table are checked every 15 minutes, and goals measured by a number on a web page every 60 minutes. Tables cost almost nothing to check, so they’re checked more often.

How shared team notes work

Shared team notes store one “label (key) and content” pair at a time. Writing to the same label again replaces the content by default (or adds to what’s there, if the employee chooses to merge) and bumps a version number; earlier versions aren’t kept separately.
When an employee or a person updates a note, other employees see the change on their next run. The notes employees see are tagged with whether a person or a team member last wrote them.
Among employees, only the employee that last wrote a note (or the manager) can delete it. But a note last written by a person can’t be deleted by any employee, the manager included. Once an employee rewrites that note, the employee-to-employee rule applies again.
People can edit or delete any note from the screen, but a save or delete only goes through if the note hasn’t changed since they opened it; otherwise it’s refused (reload the latest content and try again).
Note names can be up to 64 characters, descriptions up to 500 characters, and content up to about 32,000 bytes. Names can’t contain line breaks or control characters, and a team can’t have two notes with the same name.
Content typed on screen is saved as structured data if it reads as that format, and as plain text otherwise.

Limits of posting and withdrawing

Posting and withdrawing only add or remove a reference to the original in the team conversation. The original’s content, its public link, and access within the organization don’t change.
For withdrawal, the “owner of the original” means the employee’s owner for a canvas or web app (the user who owns that employee, or that organization’s owner), and someone who can manage the file for a file (whoever uploaded it, the person an employee delivered it to as a result, or the organization owner). When an employee made the post, that employee’s owner can also withdraw it. Being a regular organization member isn’t enough to withdraw someone else’s post.
A withdrawn post doesn’t reappear in the team’s resource list through old messages. Posting again creates a new post message.

The team work guide card in one-on-one conversations

The card only appears in a one-on-one conversation where a person is talking directly. It doesn’t appear when an employee reads team work during a scheduled run or other automatic run, inside team chat, or on externally shared pages.
The employee searches the team’s work for items whose title or summary contains every word of the search, most recently updated first, up to 5 candidates, and only shows a card once the work the person means is pinned down to one. The server never guesses which work to pick.
Only the server creates the card’s navigation buttons. An employee can’t make up a button that leads to an arbitrary team or piece of work.
Whoever picks up the new work started from a follow-up also receives a summary of the original work’s result (up to 600 characters).

How file handoffs actually work

A handoff moves the file directly from the sending AI employee’s AI work computer (the sandbox where it runs code) to the receiving employee’s AI work computer. It never passes through the backend server.
The actual transfer isn’t coordinated when the mention with the attachment is sent. It’s coordinated when the recipient’s next run starts. If the recipient never runs again, the file never arrives.
The 500MB size cap and 7-day retention window are deployment settings and may change.
A message with an attachment must have exactly one @mention, because each handoff bills AI work computer uptime on both sides. Fanning a file out to several people would multiply that cost.
Files under any hidden folder (one starting with .), where an employee’s auth tokens, long-term memory, and settings live, are excluded from what can be exported.

What’s shared is the “okay to install” decision for one skill, not a roster of what every team member has installed. Another team member still has to search for it when their own work needs it, then install and load it into their own list before they can actually use it. Nothing gets auto-installed for someone who doesn’t need it.
The approval record and what the employee currently reads as its working context (a note describing which skills are installed) are managed separately. When a conversation gets tidied up or that note gets summarized, the approval record itself stays put, so it isn’t asked again.
If the approval was for something coming from an external source like a GitHub repository, the content is pinned to what it was at the moment of first approval, and re-fetching it later is treated as the same thing. If the upstream source changes after that, the earlier approval doesn’t automatically extend to cover the new version.

How the result option is judged

The option is fixed on each part when the work is split. If the team lead doesn’t set one for a part, the assigned employee’s setting applies, and if there is none, Make a deliverable applies. Automatic checks that don’t fit the option (for example, whether a received file’s text contains a phrase, in Review) aren’t verified by the server and fall back to the team lead’s judgment. Review only accepts files whose content matches what was handed over from the earlier step, and leaves out files that changed after being handed over. Take action passes when any of that part’s email send records shows success, fails when all of them failed, and goes to the team lead’s judgment when there is no record or sending isn’t confirmed yet. The server assigns task numbers so they never collide even when tasks are created at the same moment, and the prefix is the team’s setting combined with the number when displayed.

Tool setup for AI employees

There is no “equip” step for a team employee. Teammates start with every tool (the team lead only gets the allow list above), and skills can be installed from that employee’s own settings screen at any time.

If you want to give an existing employee more skills, remember each employee is independent with its own detail page. Just install them from that employee’s Settings tab, the same way you would for any standalone employee. You don’t need to go through anyone for this. See Equipping Tools for details.

Adding a new employee to an existing team is likewise something team employees (team lead included) can’t do themselves. You hire the employee with + Hire and then move it into the team (see “Option 4” above). It’s the same single confirmation you get when hiring a standalone employee. See Handling Approvals for details.

How the team chip and back button are decided

The team chip is driven by the same team list the employee table uses. While that list is still loading, the chip stays hidden instead of flashing in and out, and it never appears for employees on the default team. Team chat is structurally blocked there, so the chip would be a door that doesn’t open.

Where the back button goes depends on how you got to that screen, not its URL. That routing information survives a page refresh, but it doesn’t travel with a copied team chat link or a link opened in a new tab. So a colleague who follows a shared team chat link always lands back in Home on their way out, never in the sender’s own private conversation.