Coworker details

What the details page shows, and what its Supervisor can change: job description, expectations, Supervisor, instructions, model, Knowledge Bases, Team, Team access, secrets and connectors.

For: Anyone who can see the Coworker; changes need its Supervisor · Last updated

Open a Coworker's details page by selecting Details on its roster card. Anyone who can see the Coworker can open the page. At the top you see its name and phase, plus Export and Chat with the Coworker's name. For a Coworker in the Interview phase, those two buttons appear only for its Supervisor. Below that are its role, practice, Supervisor and Team.

Who can change things#

A Coworker's own Supervisor can make changes on this page. Everyone else sees the same information without the edit buttons. See What your abilities mean.

A Coworker that is awaiting approval shows a Proposed badge and a banner, Proposed — awaiting approval. It has no Chat with button, no Triggers card and no terminal, and approvers get Approve to Interview, Leave feedback and Retire buttons, as described in Proposed Coworkers.

A banner at the top appears when a promotion request is waiting for approval or was declined, showing the sign-off and the decision note. Approvers get Approve to the new phase and Decline buttons on it, and the person who asked gets Withdraw request. See Advancing a Coworker and retiring it.

Supervisor#

The Supervisor is the named person accountable for the Coworker. A Coworker with no Supervisor shows the warning "This Coworker has no Supervisor." Its Supervisor sees Reassign and Clear next to the name. When there is none, admins and executives give it a Supervisor from the Coworkers roster. Choose a person from the list and select Save. Clear opens a Clear [name]'s Supervisor? dialog; select Clear Supervisor to confirm. The new Supervisor must hold the manage_coworkers ability. Once you hand a Coworker to someone else you can no longer manage it.

The job description#

The left side shows the Coworker's job description, built when it was hired, or left empty when the Coworker was created with only a name and a sentence:

  • Primary Responsibilities
  • Scope Boundaries
  • Knowledge Domain, with what it needs to know and its Knowledge Bases
  • Performance Expectations, each with its time horizon and status
  • Working Relationships

A Coworker's Supervisor can edit these parts at any time, so a Coworker created quickly can gain its full job description later. Everyone else sees them read-only.

  • Primary Responsibilities, Scope Boundaries and Working Relationships: select Add or Edit (for example Edit responsibilities), write one item per line and select Save.
  • Knowledge Domain summary: Add a summary or Edit summary.
  • Performance Expectations: Add expectations or Edit expectations. Each row has a horizon of 30, 90 or 365 days and a criterion saying what good looks like by then; each row needs a criterion.
  • Opportunity Scorecard: Add a scorecard or Edit scorecard. Score the seven dimensions from 1 to 3; Save is available once all seven are scored. The saved card shows the total out of 21 with a verdict: Strong first-deployment candidate, Viable, with reservations or Scores low — worth a closer look before investing.

The Performance Expectations, Opportunity Scorecard and Working Relationships sections start collapsed; open them to see their contents. Knowledge Bases are attached separately, as described below.

Profile image#

Each Coworker can have a profile image. It appears on the roster, in the chat header and thread list, in the Coworker picker and at the top of a new chat. Without one, the Coworker's initial is shown in a circle. Everyone who can see the Coworker sees the image; only its Supervisor can change it.

At the top of this page, the Supervisor can:

  • Select Upload image and choose a PNG, JPEG or WebP file of up to 2 MB. SVG and GIF files are not accepted, and a file whose contents don't match its type is refused.
  • Select Generate image, describe the picture, and select Generate. The result is shown as a preview and nothing is saved until you select Use this image; select Generate again for another, or Discard to keep the current image. Generating needs an image-generation model: an admin chooses one in Settings, or you can upload an image instead.
  • Select Edit alt text to change the description that screen readers read aloud. It starts as "Profile image of" followed by the Coworker's name and can be up to 200 characters.
  • Select Remove image and confirm to go back to the initial.

The image can't be changed once a Coworker is retired.

Conversation starters#

Conversation Starters are short first messages that appear as buttons on a new chat with this Coworker, under Try asking. Beside them the new chat shows What I do and What I don't do, taken from the Primary Responsibilities and Scope Boundaries above.

A Coworker's Supervisor can manage the starters; everyone else sees them read-only. Select Add starters or Edit starters, then:

  • Select Add a starter and type a message, or use Remove to delete one. You can have up to 6 starters of up to 140 characters each; blank and repeated entries are dropped when you save.
  • Select Suggest starters to have a draft written from the Coworker's role, responsibilities and scope boundaries. The draft only fills the editor: edit or remove anything, then select Save to keep it. If no default model is set, an admin can choose one in Settings, or you can write the starters yourself.

Starters can't be changed once a Coworker is retired.

Instructions#

Instructions is freeform guidance from the Supervisor. It is layered onto the job description in every answer the Coworker gives. Company-wide instructions set by an administrator apply first, and the Coworker's own instructions refine them. With no instructions, the page says "No additional instructions set."

To change them, select Add instructions (or Edit instructions), type your guidance, and select Save. Cancel discards your edit.

Knowledge Bases#

Under Knowledge Domain, Knowledge Bases lists what the Coworker draws on. This list and its buttons follow the Knowledge Bases switch on Settings, then Features; when the switch is off, Knowledge Domain shows only the summary.

  • Select Attach a Knowledge Base, choose one from the list and select Attach. Knowledge Bases marked human-only are not offered, because they cannot be given to a Coworker. A retired Coworker cannot have new Knowledge Bases attached.
  • Select Remove next to a Knowledge Base to detach it.

If a Knowledge Base is limited to certain Teams, a warning explains the effect: either nobody who can reach the Coworker can see that Knowledge Base, or some people who can reach the Coworker cannot see it and so may get different answers.

Team#

The line under the Coworker's name shows its Team, used to group Coworkers by part of the organization. Its Supervisor sees an Edit button; choose a Team, or No team, and select Save.

Model binding#

The Model binding card shows the model the Coworker runs on. Its Supervisor can choose another model and select Rebind; a "Saved" note confirms it. A retired Coworker shows Unbind model instead, which releases the model it was using.

Team access#

For a Full-time Coworker, the Team access card controls who can use it. By default everyone in the company can work with it. Its Supervisor can choose one or more Teams from the list to limit access to those Teams, plus the Supervisor. Select Save to apply or Cancel to discard. Other people see "Open to everyone." or the list of Teams.

Signals#

The Signals card gives the evidence for a promotion decision: how well people were helped, how widely the Coworker is used, and what went wrong. Only the Coworker's Supervisor and people with the admin or executive ability see it; for everyone else the card does not appear. Nothing on it is shown per person, and the list it links to for the Supervisor shows no names either.

Use 30 days or 90 days at the top right to change the period. The card shows:

  • Helped: the share of reviewed interactions where the person was helped. An interaction is a message, the Coworker's reply and the person's follow-up. A sample of them is reviewed automatically, and each is judged helped, struggled or unclear. The rate counts helped and struggled interactions only, and it appears once at least 20 of them have been reviewed. Below that the card says "Not enough reviewed interactions to show a rate." and how many have been reviewed. When automatic review is switched off for your company, the card keeps the earlier rate and its count, and adds "Automatic review is off. Earlier reviews are shown."
  • Thumbs up: the share of thumbs-up among the thumbs-up and thumbs-down ratings people gave, shown once there are at least 5 ratings. Interactions a person rated with a thumb are not reviewed a second time.
  • Helped by duty: the helped rate for each of the Coworker's Primary Responsibilities, with how many interactions were reviewed under it. A duty's rate appears once 5 of its interactions have been reviewed; before that the row shows only the count. A line under the list says what share of reviewed interactions did not match a listed duty.
  • Where people struggled: how many struggled interactions fell under each reason: Person corrected a mistake, Person asked again another way, Reply left out what was needed, Person gave up, Reply was an error or Other. For the Coworker's Supervisor, each reason is a link, and Read these interactions opens the Coworker's guided review with that reason chosen in its list, where the Supervisor can read them and mark how each was resolved. See Running a guided review. Admins and executives see the counts only.
  • Reach and usage: how many people used the Coworker in the period and, when the Coworker has an invited cohort, how many of them did. A small bar chart shows interactions per week, with the interaction total for the period below it.
  • Last activity: when the Coworker was last used.
  • Critical errors: replies that failed in a way trying again would not normally fix. Failures that clear on their own, such as a busy service, are counted separately and noted underneath as temporary errors that recovered or were retried.
  • Tool failures: how many of the Coworker's tool calls failed, with the percentage once there are at least 10 calls, and a list of the tools that failed most. A saved script shows under its own name.

A Coworker's Supervisor also sees Review new interactions at the bottom of the card, with the line "Checks interactions that haven't been reviewed yet." It reviews the Coworker's interactions from the last 30 days that have not been reviewed, straight away instead of waiting for the next automatic round. It does not open the interactions listed under Where people struggled; use Read these interactions for that. The card says "Review started. Results appear here as they finish." or, when there is nothing to review, why: every eligible interaction has already been reviewed, this week's review limit has been reached, the review share set by your admin doesn't include any new interactions, or there are no interactions in the last 30 days. If automatic review is switched off or has no model to use, the button shows the reason in plain words, and an admin can fix it in Features.

Only finished interactions are reviewed, so the card reads a little low on both helped and struggled counts. It tells you about trends, not about every interaction.

Secrets and connectors#

Its Supervisor also sees three collapsible sections:

  • Secrets chooses which secrets the Coworker can use and holds values that belong to it alone. Values can be replaced but never viewed. Each secret shows when it was Last used, which Coworkers it is Used by, and a History of changes. See Secrets.
  • Connectors lists your organization's connected systems and every connector built for your organization, with a control to enable or turn off each one for this Coworker. See Connectors for a Coworker.
  • Registered scripts lists the code the Coworker has saved to reuse across runs, showing the latest version of each. Expand a script to read its source, choose Edit to save a new version, or choose Delete to remove the script and all of its versions. The Coworker can no longer run a deleted script, and past conversations are not changed.

Memory#

The Memory section holds the short notes the Coworker keeps in mind in every conversation. Anyone can see and manage their own private memories there; the Supervisor also manages the shared ones. See Coworker memory.

Lifecycle, vitals and reviews#

  • Lifecycle lists each phase the Coworker has been through, with start and end dates and notes. Its Supervisor also gets the buttons to advance or retire it; people with the admin ability see the advance button only. When your company requires approval or measures for a move, the advance dialog shows them and a request can wait for approval, as described in Advancing a Coworker and retiring it.
  • Vitals shows the count of thumbs up and thumbs down feedback, and this month's spend. When some of this month's calls have no known price, the spend leaves them out and shows how many are unpriced.
  • Reviews lists each recorded review with its date, outcome and who recorded it. Its Supervisor also sees Start a review. See Reviews and feedback.
  • Triggers lets people who can work with the Coworker set up work the Coworker does on its own. It is not shown while the Coworker is awaiting approval. See Triggers.

Terminal#

The Supervisor of a Coworker that is neither retired nor awaiting approval also sees a VM Terminal card. Connect to terminal opens a live shell on the Coworker's computer, and Back closes it.

Exporting a Coworker#

Select Export to download the Coworker's complete record as one file: its details and job description, Knowledge Bases, lifecycle history and feedback. Anyone who can use the Coworker can export it; for a Coworker in the Interview phase, that is its Supervisor.