AI agent: three layouts to pick from
Usability round, step 3. The AI agent area is six pages: Behaviour, Knowledge, Playground, Settings, Log and Triggers. Each option shows Behaviour at 1440 and Settings on a phone, in the real shell, on the six text roles. Pick one, or name the parts you want from each.
What is off now
From the apple-design review of the six pages (shots of 2026-09-27 at 1440 and 390).
- HighNothing on the page shows the other five AI agent pages. At desktop width they are only in the menu that opens from the side rail.
- HighBehaviour puts a card inside a card inside a card, and "Role" and the heading under it are nearly the same size. The text runs 460 px wide inside a 1000 px card.
- HighSettings is one 5,300 px page (10,000 px on a phone). Each message sits in a box that ends "Nobody has written this. It is the standard wording." in the same ink as the message the customer gets.
- HighSettings shows internal names (
agentReview.threshold) and switches between "the agent", "the AI" and "the bot". - MediumPlayground's Test and Manage switch is about 29 px tall on a phone, where 44 is the size a thumb expects.
- MediumKnowledge shows four text documents as cards with a "Rows per page 24" control.
- MediumLog writes "9/27/2026, 12:41:39 PM" where Knowledge writes "Sep 27, 2026", and "Published by Admin" looks like a button.
- MediumTriggers says "Standing down" and "Still using the list above" under an empty list.
Tabs and a page of text
The six pages as tabs under the area's name. Behaviour reads like a document: one heading per section, the text in a readable column, and an outline at the side that says which parts are yours.
1440 · Behaviour, Conversations
390 · Settings
- No cards on Behaviour: sections are separated by a line and space.
- The outline jumps to a section and marks it "Standard" or "Yours".
- Settings keeps one long page, but each message shows once, with "Standard wording" as a small status under it.
- Trade-off: Settings is still long; the outline helps on desktop only.
A sidebar and a list
A sidebar for the AI agent area, with the four situations under Behaviour. Settings becomes a grouped list: each row says what the setting is now, and opens its own page.
1440 · Behaviour, Conversations
390 · Settings as a grouped list
- One card per heading, never a card inside a card.
- The situation you are in is in the sidebar and in the page title.
- Settings fits on one phone screen: every row shows its current value.
- Trade-off: 220 px go to the sidebar; editing a setting is one more tap.
Rows that open a sheet
Behaviour lists its five headings as rows: the first words of each, and whether it is standard, yours or waiting to be published. A row opens beside the list to read and edit. Settings splits into four short pages.
1440 · Behaviour, Goal open beside the list
390 · Settings, the replies page
- The whole situation fits one screen; you read one heading at a time.
- Edits go to the draft as today; the row says "Draft" until you publish.
- Settings: Replies, Audience, Thresholds, Moderation, each a page you can finish reading.
- Trade-off: you no longer see the whole prompt at once; "See the exact prompt" still shows it.
In every option
- The heading drawn as "Who the AI is" waits on your answer about the stored prompt, which still says "Who the bot is". Labels such as "When the bot gets stuck" become "When the AI gets stuck".
- Knowledge becomes a table: title, where it applies, facts, on or off, last change. No pager under 25 documents.
- Dates read the same everywhere: "27 Sep, 12:41", with the year only when it is not this year.
- Log names who and what in one line: "Admin published Conversations · Tone (added)".
- Playground's Test and Manage switch is 44 px tall on a phone.
- Internal names such as
agentReview.thresholdleave the screen. - The standard messages drawn here are the current defaults, which no longer carry a dash ("…on our side. A Silver Oak consultant…").