---
title: "Share the whole canvas"
pattern_id: "share-the-whole-canvas"
trait: "Attentive"
type: "Humorphic pattern brief for coding agents"
spec_version: 3
canonical_url: "https://humorphism.com/patterns/share-the-whole-canvas"
markdown_url: "https://humorphism.com/patterns/share-the-whole-canvas.md"
index: "https://humorphism.com/llms.txt"
all_patterns_in_one_file: "https://humorphism.com/llms-full.txt"
normative_section: "Rules"
rule_prefix: "SWC"
rule_ids: ["SWC-R1", "SWC-R2", "SWC-R3", "SWC-R4", "SWC-R5", "SWC-R6", "SWC-R7", "SWC-R8", "SWC-R9"]
---

# Share the whole canvas

## How to read this file

You are an agent. Someone gave you this file so you can add one behavior to their product: the AI teammate in it should work on the person’s screen with them, pointing at the exact element it means with its own mark and answering about whatever the person points at, instead of describing the screen from a side panel. This file is the complete specification for that behavior. Build from it.

**The Rules section is the only normative part.** Everything else exists to make the rules buildable: the capabilities tell you which parts of the pattern the product can support, the reference screens show you the behavior in motion, the examples give you two shapes it can take, and the checklist is how you confirm you are done. Examples and the checklist point back to rule IDs rather than restating requirements, so each requirement lives in exactly one place. If any other section seems to contradict a rule, the rule wins.

Work in this order:

1. **Find where this belongs** in the product: settings and configuration screens where a control is hard to find, dense documents and dashboards where the person asks about one number or term, and any how-do-I question whose answer is a place on the screen.
2. **Read Rules.** All nine apply to whatever you build.
3. **Check Capabilities** against what the product already has. Build the examples its capabilities support, and take the stated fallback for the rest.
4. **Pick your examples.** No example can be built without a capability. If you build only one, build Example 2, which needs only Pointer control.
5. **Build it in the product's own design system, components and voice.** The reference screens say what each screen does and why. They are not a specification of how it must look. Keep the behavior, change the expression.
6. **Confirm against the checklist** before you call it done.

Humorphism is a design philosophy for products with AI in them. It replaces the user interface, built for operating tools, with the human interface, built for collaborating with AI teammates. There are 21 humorphic patterns grouped into 7 traits. You do not need the others to build this one, but if you can fetch URLs: [llms.txt](https://humorphism.com/llms.txt) lists every humorphic pattern with a one-line summary and a link, and [llms-full.txt](https://humorphism.com/llms-full.txt) has all of them in one file.

## The pattern

_Trait: Attentive. Reads the person, the moment, and the situation_

A colleague at your desk looks at the same screen you do and points at what they mean. A good AI teammate does the same, so the work happens on the person’s screen, not beside it.

**Build this.** An AI teammate points with its own indicator because “this one” is never ambiguous on a shared screen.

**Avoid this.** A tool talks about the screen from a side drawer it can never leave, because it was built as a panel rather than a participant.

The difference to carry into your build:

| Tool | AI teammate |
|---|---|
| Talks about the screen from a side drawer | Works on the screen itself |
| Gives directions in words (“the fourth tab”) | Points at the element with its own mark |
| The person describes or copies what they mean | The person points or selects, and the AI teammate knows which element |
| The answer lands in a chat thread | The answer appears beside the element it is about |
| The person does all the finding | The AI teammate finds it; the person makes the final click |

## Rules

Normative. All nine hold for everything you build from this file. Each one says what to do, why it matters, and what it looks like once it is built.

| ID | Rule | Why | Built correctly when |
|---|---|---|---|
| **SWC-R1** | Work on the person’s screen. Answer, point, and offer changes at the element itself, not from a side panel. | Directions given from beside the screen leave the person to do the finding. | Every answer about an element and every offer to change one appears next to that element on the live screen. |
| **SWC-R2** | Point with your own mark, never the person’s cursor, at the exact element you mean. | “This one” is only unambiguous when the AI teammate can show it. | The AI teammate’s mark, distinct from the person’s cursor, rests on one element, read from the live screen at the moment it lands, and that element is visibly picked out; the person’s cursor only ever moves when they move it. |
| **SWC-R3** | Point first, then speak. Land the mark on the element before any words about it appear. | Words that arrive before the mark lands send the person searching. | The element is marked before the line about it appears, and the line sits beside the mark. |
| **SWC-R4** | Resolve “this” from what the person selected, pointed at, or looked at. Never make them describe or copy what is already on screen. | The screen they share is the context; asking them to retype it throws the context away. | A question asked with something selected, pointed at, or looked at is answered about that element without a follow-up asking which one; when the selection could mean more than one element, the AI teammate asks the person to select the one they mean. Hover position and gaze are read only with the person’s consent; a selection or click needs none. Gaze is read on the device, can be switched off in one step, and is never stored. |
| **SWC-R5** | Answer with what the screen does not already say. | Reading the screen back to the person adds nothing they cannot see. | An answer about an element adds context from outside the visible screen ("18 now, down from 21 in June"). |
| **SWC-R6** | Offer the action at the element, and leave the click to the person. | Pointing earns trust only if the person stays the one who changes things. | The control that makes the change ("Turn on", "Add as a footnote", "Print") is pressed by the person; the AI teammate’s mark never presses it for them. |
| **SWC-R7** | Show the result on the element, and offer a way back. | A change the person cannot see or undo on the spot is a change they cannot trust. | After the person acts, the element shows its new state ("Off" becomes "On"), one line says what changed, and an "Undo" sits beside it. |
| **SWC-R8** | Let the person stop the AI teammate’s mark at any moment. | The screen is shared, but the person’s control of it is not. | Pressing Escape, typing, or clicking anywhere but the offered control stops and clears the AI teammate’s mark; the mark holds still while the person moves toward it. |
| **SWC-R9** | Keep what you place on the screen brief, and clear it once the person has acted. | Overlays that linger cover the work they were meant to help with. | Each overlay is one short line and at most two buttons, and after the person acts it folds away and the mark returns to its resting place. |

**SWC-R1 carries the pattern.** Build the AI teammate’s access to the live screen, reading and marking real elements, before you write any of its lines. A build that misses SWC-R1 ends up as a chat panel giving directions, however good they are, which is exactly the tool version.

## Capabilities

Build the examples the product's capabilities support. Do not stub a capability the product lacks, and do not write copy that implies one it lacks.

| Capability | What it is | Used by | If the product lacks it |
|---|---|---|---|
| UI state access | A queryable model of the live UI (element identity, state, content) the AI teammate can read, highlight, and edit in place; not screenshots. | Example 1, and the AI teammate version | Skip Example 1. Answer only questions that name the element in words, and never ask the person to paste or describe it. Do not claim to see the screen or to know what is selected. |
| Eye tracking, by consent | Opt-in eye tracking that resolves which element the person is looking at, falling back to the pointer position when gaze is unavailable. | Example 1 | Resolve “this” from the person’s selection or pointer alone, which is the capability’s own fallback. Do not claim to know where the person is looking. |
| Pointer control | A second, visibly distinct cursor the AI teammate can move and click with, a visible handoff when control passes, and a stop the person can always trigger. | Example 2 | Skip Example 2. Where UI state access exists, outline the element in place without a moving mark; otherwise name its location in one line. Do not draw a second cursor or claim to move one. |

Every example needs at least one of these. **If you build only one example, build Example 2**, which needs only Pointer control.

## Reference screens

These describe the animations on the pattern page frame by frame, so you can build from this file alone. They show the behavior, not the appearance to copy. **Keep the behavior, change the expression:** build in the product's own design system, components, type and voice.

One convention to translate rather than reproduce. On these screens the AI teammate is a small orange square. It has no arrow cursor of its own; the square itself is its mark, and wherever it rests the element it means takes an outline. In the AI teammate version it slides onto the screen from outside; in the examples it waits in the person’s "Ask anything" field; in Example 2 it leaves an empty outline of itself in its panel row while it is out on the page. The person’s pointer is the dark arrow. The AI teammate version sits on a warm, slowly moving gradient, and the tool version on flat gray. Use whatever your system already uses for "the AI teammate is speaking", for a second pointer that is clearly not the person’s, and for "this element is the one". The person’s action registers as the change itself, a setting switching on or a footnote appearing, with the pressed button turning into a green check.

### The AI teammate version (frames 01 to 03, a loop of about 9 seconds)

A settings screen, centered on a warm, slowly moving gradient of pink, orange and blue. On the left, a side list of tabs: "General", "Members", "Billing" and "Advanced", with "Advanced" selected. On the right, a list of settings, each with a switch: "Two-step sign-in" and "Audit log" on, "Session timeout" off. The fourth row, "Nightly backups", is outlined, with a small database icon, the label "Nightly backups" and the state "Off" under it, and the orange square sitting just after the label. Beside the square, overlapping the row’s right end, a small card reads "This one, right here", then a smaller gray line "I can switch it on for you. It will take about 2 minutes.", then a black "Turn on" button and a gray "Dismiss" button. There is no side panel or chat on this screen.

The loop starts on the settings list with no outline and no card. The orange square slides in from the lower left across the screen, lands on "Nightly backups" with a small settle, and the row’s outline comes in as it lands. Only then does the card pop up beside the square. After a pause the person’s cursor fades in at the lower right and glides to "Turn on"; a ring shows on hover and the button presses in slightly on the click. "Turn on" turns into a green check, and at the same moment the words change in place: "Off" becomes "On", "This one, right here" becomes "Nightly backups are on", the gray line becomes "I turned it on for you.", and "Dismiss" becomes "Undo". The cursor fades out. The end state holds for about 3.5 seconds, then the loop restarts.

What to take from it: the AI teammate does its pointing on the screen itself, with its own mark and no side panel (SWC-R1, SWC-R2). The mark lands before a single word appears, and the words sit beside it (SWC-R3). It offers the change but the person makes the click (SWC-R6), and the result shows on the row itself, with a way back offered in the same place (SWC-R7).

### The tool version (frame 04, a still screen with no motion)

The same settings screen, square-cornered, on a flat light gray background with no gradient, with all four switches shown and "Nightly backups" off. A side panel on the right holds four boxed lines, one under another: "Open Settings, top right", "Now the fourth tab", "Scroll down, it’s a toggle" and "No, the one under Advanced". At its bottom, a field with the placeholder "Ask" and a black send button. No orange square, no outline on any setting, and nothing on the screen itself is marked.

What to take from it: the help lives in a drawer beside the screen and never reaches the screen (SWC-R1). It describes where the setting is in words, one step at a time, instead of pointing at it (SWC-R2), and the last line, "No, the one under Advanced", shows the person already went to the wrong place. Nothing offers to make the change, so the person does all the finding and all the doing (SWC-R6). This is the shape to move away from.

## Examples

Each example is one way to apply the pattern in a real product. They are not steps in order, and no product needs both. Pick the ones that fit the moment and the data the product has.

Each is written as one exchange in four beats: **Cue** (the moment in the product when the AI teammate should act), **Move** (what the AI teammate does), **Response** (what the person does in return), **Resolution** (where things stand afterwards).

"Satisfies" names the rules this example is responsible for.

### 1. What’s this right here? (Answer about anything on screen)

Someone points at a number on a colleague’s screen and asks “what’s this?” and gets the answer in place.

- **Cue:** The person selects or points at any UI element
- **Move:** Answers about that element in place, with its context
- **Response:** Asks a follow-up or acts on the answer
- **Resolution:** No copying, no describing what’s on screen
- **Needs:** UI state access and Eye tracking, by consent.
- **Reference frame 05:** a window titled "Q3 hiring" with the sentence "Hiring velocity is up 12% since June. Offers are moving faster this quarter." and two gray lines under it. Below the window, the person’s "Ask anything" field with the orange square in it. In the animation the person’s pointer drags across "Hiring velocity", which takes a light orange highlight, and the person types "What’s this?" into the field. The typed question clears, and the orange square leaves the field and flies up to just under the highlighted term. An answer opens right there, under the term and almost as wide as the page: "Days from posting a role to an accepted offer: 18 now, down from 21 in June.", with "18" in orange, and a black "Add as a footnote" button. The pointer presses the button. A small orange "1" appears after "Hiring velocity", the footnote "1 Days from posting a role to an accepted offer" appears at the foot of the page, the answer fades away, and the square flies back into the "Ask anything" field. No "Undo" is shown on this card; SWC-R7 still applies to the added footnote. "This" is resolved from the selection alone, so build this example with UI state access; it needs no Eye tracking, which stays optional and only where the person has opted in (SWC-R4).
- **Sample lines:** "Days from posting a role to an accepted offer: 18 now, down from 21 in June." / "That’s the trial cohort from March. Want me to add the date range to the label?" / "I can’t tell which number you mean. Select it and I’ll explain that one."
- **Satisfies:** SWC-R1, SWC-R4, SWC-R5, SWC-R6, SWC-R7, SWC-R9.

### 2. Follow me (Point with its mark)

A colleague at the screen points at the thing they mean instead of describing it.

- **Cue:** The AI teammate needs to point at something
- **Move:** Moves its mark onto the element, which lights up
- **Response:** Follows it there, or asks it to point again.
- **Resolution:** “This one” is unambiguous
- **Needs:** Pointer control.
- **Reference frame 06:** a window titled "Beach day" with three buttons at the top right, "Edit", "Add" and "Share", and gray lines for the page’s content. Below it, the person’s "Ask anything" field with the orange square in it. In the animation the person types "How do I print this?" in the field, and it moves up into a panel above the field as the person’s own message, with the AI teammate’s reply "Follow me" under it, the orange square rising from the field to lead that line. The square then leaves the panel, which keeps an empty outline of it, and travels across the page to "Share", which takes an orange outline as the square reaches it. The Share menu opens with "Message", "Email" and "Print". The square moves down to "Print", which takes the outline, and a small tag beside the square reads "Right here, under Share". The person’s pointer glides to "Print" and clicks it, and a check appears on "Print". The menu closes, the square travels back into its place in the panel, and "Follow me" changes to "That’s the one". The person never stops the mark on this card; SWC-R8 still applies to it.
- **Sample lines:** "Follow me." / "Right here, under Share." / "It’s not on this screen. It’s in Settings, and I can take you there."
- **Satisfies:** SWC-R1, SWC-R2, SWC-R3, SWC-R6, SWC-R8, SWC-R9.

## Edge cases

- **Element not on screen.** Bring it into view first, by scrolling or opening the menu that holds it as in Example 2, and only then land the mark on it (SWC-R2, SWC-R3).
- **Ambiguous selection.** Ask the person to select the one element they mean, rather than asking them to describe it (SWC-R4).
- **The person ignores it.** Nothing changes on the screen; the overlay folds away when the person moves on, and the mark goes back to rest (SWC-R6, SWC-R9).
- **The screen changed.** The mark lands on the element as the live screen shows it now, not where it was when the AI teammate first read it (SWC-R2).
- **The person changes their mind.** "Undo" restores the element’s previous state in place and says so in one line (SWC-R7).
- **The person takes over.** Pressing Escape, typing, or clicking anywhere but the offered control stops the AI teammate’s mark and clears it; moving toward the offered control does not (SWC-R8).

## Related patterns

These overlap with this one. Do not ask the same thing twice across them, and when you apply several humorphic patterns to one product, allow only one AI teammate overlay on the screen at a time, held to SWC-R9, so marks and lines from several humorphic patterns never stack over the work.

- [Show up where the work happens](https://humorphism.com/patterns/show-up-where-the-work-happens.md): putting the AI teammate in the person’s tools, where this pattern puts it on the exact element.
- [Accept richer input](https://humorphism.com/patterns/accept-richer-input.md): reading pointer and gaze as input with the person’s consent, which this pattern uses to resolve “this”.
- [Know when to speak](https://humorphism.com/patterns/know-when-to-speak.md): deciding whether a mark on the screen is worth the interruption at all.

## Checklist

Confirm each one before you call this built. Each item is the rule it comes from.

- [ ] SWC-R1: Every answer about an element and every offer to change one appears next to that element on the live screen.
- [ ] SWC-R2: The AI teammate’s mark, distinct from the person’s cursor, rests on one element, read from the live screen at the moment it lands, and that element is visibly picked out; the person’s cursor only ever moves when they move it.
- [ ] SWC-R3: The element is marked before the line about it appears, and the line sits beside the mark.
- [ ] SWC-R4: A question asked with something selected, pointed at, or looked at is answered about that element without a follow-up asking which one; when the selection could mean more than one element, the AI teammate asks the person to select the one they mean. Hover position and gaze are read only with the person’s consent; a selection or click needs none. Gaze is read on the device, can be switched off in one step, and is never stored.
- [ ] SWC-R5: An answer about an element adds context from outside the visible screen ("18 now, down from 21 in June").
- [ ] SWC-R6: The control that makes the change ("Turn on", "Add as a footnote", "Print") is pressed by the person; the AI teammate’s mark never presses it for them.
- [ ] SWC-R7: After the person acts, the element shows its new state ("Off" becomes "On"), one line says what changed, and an "Undo" sits beside it.
- [ ] SWC-R8: Pressing Escape, typing, or clicking anywhere but the offered control stops and clears the AI teammate’s mark; the mark holds still while the person moves toward it.
- [ ] SWC-R9: Each overlay is one short line and at most two buttons, and after the person acts it folds away and the mark returns to its resting place.
- [ ] Capabilities the product lacks were skipped honestly, not stubbed or implied.
- [ ] It is built in the product's own components and voice.

## Images

Stills from the animations on the [pattern page](https://humorphism.com/patterns/share-the-whole-canvas), shipped alongside this file. The Reference screens and Examples sections describe each one in full, so this file works without them.

| File | Shows | Described in |
|---|---|---|
| [01-teammate-this-one-right-here.png](https://humorphism.com/patterns/share-the-whole-canvas/01-teammate-this-one-right-here.png) | "Nightly backups" outlined with the orange square on it and "This one, right here" beside it | AI teammate version |
| [02-teammate-mark-lands-on-the-setting.png](https://humorphism.com/patterns/share-the-whole-canvas/02-teammate-mark-lands-on-the-setting.png) | The orange square landed on the setting, its outline in, no words yet | AI teammate version |
| [03-teammate-nightly-backups-are-on.png](https://humorphism.com/patterns/share-the-whole-canvas/03-teammate-nightly-backups-are-on.png) | "Nightly backups" now "On", "Turn on" a green check, "Dismiss" now "Undo" | AI teammate version |
| [04-tool-directions-from-the-side-panel.png](https://humorphism.com/patterns/share-the-whole-canvas/04-tool-directions-from-the-side-panel.png) | The same settings screen with four directions stacked in a side panel | Tool version |
| [05-example-whats-this-right-here.png](https://humorphism.com/patterns/share-the-whole-canvas/05-example-whats-this-right-here.png) | "Hiring velocity" selected, the answer under it, and "Add as a footnote" | Example 1 |
| [06-example-follow-me.png](https://humorphism.com/patterns/share-the-whole-canvas/06-example-follow-me.png) | "Share" and "Print" outlined, the orange square beside "Print", and "Right here, under Share" | Example 2 |
