Activity shows everything an agent finished

What changes

Activity today lists only five kinds of work, each wired up by hand, and it started recording on 5 October. Anything newer than that list — a page Snoopy creates, a page it archives, a Plane task, a routine it sets up, a channel setting it changes — never shows, and a routine's daily brief only shows because the agent wraps its answer in a special format.

With this change, every write an agent finishes is recorded by the write itself. Activity reads that record. There is nothing for an agent to remember and no list to keep up to date: when we add a new kind of write next month, it shows up in Activity on the day it ships. Past agent work is filled in from the agents' run records on launch day, so channels don't start empty.

Each row says who the agent did it for, not which agent did it: there is one main agent today, so its name is left off. When there are several, it can come back on the bottom line.

Tapping a row

Everything that can appear

One row per thing an agent changed in a turn. The icon says what kind of thing changed; the title names it, using the name it had at that moment; the subtitle says who it was for.

Never in Activity: ordinary answers that saved nothing, questions the agent asked, approvals still waiting, failed attempts, and anything a member did by hand.

What our workspace would have shown

These counts are the agent writes in our own workspace over the 14 days to 5 October, read from the agents' run records. The channel names are ours; nothing else from those runs appears here.

"Up to" because a channel creation waits for someone to approve it; only approved ones become entries. Today's tab actually holds one entry across the whole workspace, because it only started recording on 5 October. With the event log, everything in the right-hand column is there on launch day, filled in from the run records.

Edge cases, with Vee's answers

1. Several writes in one reply

Nok drops meeting notes; Snoopy files them (which updates two pages) and creates a new page for the follow-up. The filing and the pages it updated count as one outcome; the new page is its own row. Vee: separate rows.

2. The same page changed again and again

Snoopy updates the Q3 budget page in five separate replies through the afternoon. Vee: every update is its own row. Within a single reply, everything done to one thing is one row: if Snoopy creates a page and then edits it before replying, that is "Created page Q3 budget".

3. Something that needed approval

Nok asks Snoopy to create #reimbursements. Nothing appears while the request waits. When Nok approves, the entry appears; tapping the row opens Snoopy's reply after the approval (or the request card, until that reply exists). A declined request never appears. Vee: no need to say who approved. Today only the person who asked can approve, so the subtitle already names them.

4. Routines

A routine's result appears once, with the clock icon. A routine run that failed, or that only asked a question, does not appear. Vee: agreed.

5. A Plane task

Vee: one row, with the cycle in the subtitle. When Snoopy files the task and adds it to the cycle in the same reply, that is one row. If it files the task now and adds it to the cycle in a later reply (for example after asking "add it to this cycle?"), those are two rows, each opening its own reply, because they happened as two separate steps in the chat.

If Plane is later disconnected, the title still opens Plane; it may ask the member to sign in there.

6. Things that changed name or went away

The title keeps the name from the moment it happened, because it is history. Vee: no need to show the current name, and an archived page gets nothing to open. A deleted channel's title is plain text too. Tapping the row still opens the agent's message.

7. The agent wrote, but its message never posted

Rare: Snoopy creates a page and then the run breaks before it can reply. The page exists, so the entry appears. Vee: show it; tapping the row says "Original message is unavailable". The title still opens the page.

8. The agent asked a question

Vee: asking a question isn't activity — and it never is. Only something the agent changed becomes a row. If Snoopy creates a page and then asks Nok which template to use, the page is the row; tapping it opens the question, because that is Snoopy's only message in that reply.

9. A channel with nothing yet

A channel where no agent has ever changed anything shows the empty state. Vee: no date.

10. What members do themselves

Vee: only agent activity. When Nok renames a page by hand, it does not appear in Activity. We still keep a private record of it (nobody sees it), so features like "Who changed this?" or Undo can use it later.

11. Failures

A write that failed (for example, someone was removed from the workspace mid-reply) never appears in Activity. The member already sees the agent's reply explaining it. Vee: OK.

12. Long names and Thai

Vee: two lines at most for the title and the subtitle; long-press on mobile (hover on desktop) shows the full text.

Entries about a private channel are hidden from members who aren't in it, as Activity does today. There is nothing to design for that, so it is no longer an edge case here.

Where the event log can go next

The event log is a record of every change in the workspace: who did it, for whom, to what, and from which message. Activity is the first thing to read it. These are directions it makes cheap, not commitments.

Vee's answers, and what's left

Decided by Turbo for the build (6 October), open to Vee's refinement later:

1. Two tap targets per row, and long-press on mobile (Tapping a row).
2. The subtitle sentence for each kind, as drafted (Everything that can appear).
3. A routine's result names the person who set the routine up.