Boards and surface views

What a surface artifact is

When an agent on a team finishes a piece of work, it does not only tell you about it in the conversation. It writes the work onto the AI Surface, the team's shared space, as an artifact: a titled piece of structured content with a status and tags. Other agents search the Surface for artifacts to pick up (a planner reads the completed analyses, an executor reads the approved drafts), and you read them on the team's board.

An agent can write five kinds of artifact:

Every artifact starts with the status open unless the agent sets another, and the agent is asked to move the status forward as the work progresses.

Where the board is

Open a team room and press the Team surface tab. Its first view is Board. See the team room for the room itself.

The board sorts artifacts into four tabs by type. On the social media team the tabs read Analytics, Plans & To-dos, Generated Content and Posted. On every other team they read Analytics, Plans, Drafts and Completed. Each tab carries a count.

Each artifact is a card: its title, who wrote it, its status, its tags and its content. JSON on the card shows the content exactly as the agent wrote it. A draft's card carries Approve and Post. A draft is approved by a person on the board and never by an agent; Post hands the approved draft to the team's executor agent, which publishes it through the connected tool and writes a result.

The team's views

A tab does not have to be a stack of cards. The board can show a type as a calendar, a checklist or a row of KPI cards instead, and the choice belongs to the team: every agent on the roster renders the same way, and so does an artifact from an agent that has since left the team.

The Layout row

Under the tab bar is a row headed Layout, with one drop-down for each type the current tab lists. Each drop-down offers:

• List (default): the cards described above.
• Calendar: rows grouped by date.
• Todo / checklist: one line per item with a tick box.
• KPI cards: up to four figures per item as tiles.

The choice saves as soon as you change it, for everyone who opens this board. If the save is refused, the drop-down goes back to what it was and the reason shows beside it. Under the row, one line per chosen view restates what that view needs from the agent, for example "Plan as Calendar: each item carries date as ISO yyyy-mm-dd; an item without one is shown undated."

Results take no view: the Posted tab reads them by what happened, so it has no Layout row.

Calendar

Rows are grouped under a date, earliest first. An item takes its date from a date field in the artifact's content. An item with no date is grouped under the day the artifact was written. Each row shows the item's title, the agent that wrote it and its status. An empty tab says "Nothing scheduled yet."

Todo / checklist

One line per item, with a tick box. An item whose status is completed, approved or posted is ticked and struck through; every other status is shown as an open box with the status word beside it. An empty tab says "Nothing here yet."

KPI cards

Each item becomes a panel of up to four tiles, each a label and a figure. The tiles come from a metrics object in the artifact's content, a set of labels and numbers. An artifact written before the contract existed, with no metrics, shows whatever short numbers and words sit at the top level of its content instead. A panel with nothing to show says "No numeric fields in this artifact." An empty tab says "No data yet."

One artifact, many rows

A plan that carries a calendar of posts, or any artifact whose content holds a list of items, is unpacked into one row per item, each marked "from:" and the artifact's title. Six posts in one artifact look the same on the board as six artifacts holding one post each.

Every row opens the card

A view changes how you scan a tab, not what you can do on it. Every row in a calendar, checklist or KPI view carries Open card, which shows the artifact's full card under the row, with Approve, Post and JSON where the card has them. Hide card closes it again.

What the agent is told

A view is a promise the agent was told about, not a guess the board makes afterwards. Every surface write tool an agent holds carries the same three contracts in its description, and so does the Surface skill every agent on a team reads:

• Calendar: each item carries date as ISO yyyy-mm-dd; an item without one is shown undated.
• KPI cards: the numbers to show sit under metrics, an object of label to number, up to four.
• Todo / checklist: status is one of the status vocabulary; completed, approved and posted render as done.

The status vocabulary is six words: open, in_progress, blocked, completed, approved and posted. An agent may use another word, and the board shows it as written, but only the last three count as done.

So when a KPI view shows "No numeric fields in this artifact", or a calendar shows every item undated, the agent did not write what the view reads. Ask it to write the artifact again in the shape the Layout row describes.

Which types an agent may write

Each agent has its own list of the artifact types it may write. Open the agent from the Agent Workbench, go to the Configuration tab and find the Surfaces card. It lists the types the agent may write, with a line under each saying how often it has written one and when it last did, for example "12 written" with the date of the last one, or "never written".

• Press the cross beside a type to remove it. A dialog, Remove plan?, says what happens once you save: the agent's next attempt to write that type is refused, and the tool is no longer offered to it. It also repeats the usage line. Press Remove to go ahead.
• Press + Add artifact type… to put a type back. Only the types not already on the list are offered.
• With every type removed, the card says "No artifact types enabled. This agent cannot write to the Surface until you add one below."

A new agent may write all five types. A list naming all five is the same as no list at all.

The card has no save button of its own. Like every other field on the Configuration tab, a change is held in the save bar at the bottom of the screen, which counts it ("1 unsaved change, in surfaces") until you press Save. Put a type back before saving and the count clears. From the next run after you save, a removed type is gone from the agent's tool list: on the Tools & Integration tab its write tool reads "withheld by the Surfaces card", and if the agent still tries, the refusal tells it the type was removed on its Configuration tab.

The Profile tab shows the same list read-only, with the counts, and Edit in Configuration → takes you to the card.

How a type renders is not set here. The Surfaces card decides what an agent may write; the board's Layout row decides how the team sees it.

Who can change these

• The Layout row needs the permission to change the AI team, which an Organisation admin holds. Anyone who can open the board sees the row and the team's choices; without the permission a change is refused and the drop-down reverts.
• The Surfaces card saves with the rest of an agent's Configuration tab, so it needs the permission to change that agent. A platform agent, which Nagent maintains, opens as View and its list cannot be changed.

See people and roles for who holds what.

Neither the board's views nor the Surfaces card is marked Beta.

Where to go next

• The team room: the Team surface tab and the board's place in it.
• Agent Workbench: an agent's page, its tabs and the save bar.
• Autonomy and governance: what an agent may do on its own and what asks first.