---
title: "Earn autonomy"
pattern_id: "earn-autonomy"
trait: "Empowering"
type: "Humorphic pattern brief for coding agents"
spec_version: 3
canonical_url: "https://humorphism.com/patterns/earn-autonomy"
markdown_url: "https://humorphism.com/patterns/earn-autonomy.md"
index: "https://humorphism.com/llms.txt"
all_patterns_in_one_file: "https://humorphism.com/llms-full.txt"
normative_section: "Rules"
rule_prefix: "EA"
rule_ids: ["EA-R1", "EA-R2", "EA-R3", "EA-R4", "EA-R5", "EA-R6", "EA-R7", "EA-R8"]
---

# Earn autonomy

## 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 take on more only as its track record earns it, by asking the person, and step back when the person corrects it. 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 three 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: approval queues where the same item comes back again and again; recurring tasks the person signs off without changes; and the moment right after the person reverts or overrides something the AI teammate did.
2. **Read Rules.** All eight 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 1, which needs only Autonomy policy.
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: Empowering. Preserves and expands meaningful human judgment_

New colleagues are trusted with more as they get things right, and so is an AI teammate. Autonomy grows from a track record rather than a setting.

**Build this.** An AI teammate notices a dozen unchanged approvals and asks to take the task over, because autonomy granted by agreement is autonomy the person controls.

**Avoid this.** A tool asks are you sure on week ten as on week one, because its autonomy is a setting and never a track record.

The difference to carry into your build:

| Tool | AI teammate |
|---|---|
| Autonomy is a setting | Autonomy grows from a track record |
| Asks “are you sure” on week ten as on week one | Notices the run of unchanged approvals and says so |
| Takes on more when a setting changes | Asks to take on more, and waits for a yes |
| All or nothing | One task or area at a time, at a level the person picks |
| A reversal changes nothing | A reversal sends it back to asking first |

## Rules

Normative. All eight 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 |
|---|---|---|---|
| **EA-R1** | Base every offer on the track record, and name it. | Autonomy earned from a record is something the person can check; autonomy from a setting is not. | An offer to take on more appears only after a run of unchanged approvals, and states that run ("You’ve approved these invoices for 3 weeks straight with no changes"). |
| **EA-R2** | Ask to take the task over. Never take it on its own. | Autonomy granted by agreement is autonomy the person controls. | The offer arrives as a control left off, and nothing changes until the person turns it on. |
| **EA-R3** | Widen one task or one adjacent area at a time. | Small steps keep each grant matched to what the record actually shows. | Each offer covers a single task or area, never a judgment call, such as a hiring or refund decision, or a decision with legal, safety, or reputational stakes, and comes only after a clean run in that task or in the area next to it. |
| **EA-R4** | Let the person set how far it goes: off, ask first, or tell me after. | A yes-or-no grant forces the person to give more than they want or nothing at all. | Each area shows its current level, the person can move it to any level directly, and the product's own levels map onto these three, with no silent level, since EA-R5 requires a report. |
| **EA-R5** | Make the grant change what happens next. | A grant that leaves the queue as it was is only a setting with friendlier copy. | Once granted, those items stop waiting on the person, the AI teammate handles them, and it reports at the level the person chose; an item outside what was granted comes back to the person. |
| **EA-R6** | Step back after a correction, say so in one line, and leave a way to restore it. | Corrections stay cheap only when they are never argued. | When the person reverts or overrides an action, that area goes back to asking first, and one line says so ("I’ll ask before I touch invoices again."), with a way to restore it. |
| **EA-R7** | Keep every grant visible, and let the person take it back at any time. | A grant the person cannot find becomes autonomy they no longer control. | Each grant is saved where the person can see it ("I saved it to your preferences"), and turning it off works at once. |
| **EA-R8** | Keep the offer to one line and one control, and make it once. | An offer that keeps coming back is pressure, not a question. | The offer is a single line with a single control, and a declined offer does not return until the record has grown. |

**EA-R1 carries the pattern.** Build the record first: count each approval, change, and reversal per task, and drive every offer from it. A build that satisfies every rule except EA-R1 has produced an autonomy settings page with a friendlier voice, which is exactly the tool version, autonomy as a setting rather than a track record.

## 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 |
|---|---|---|---|
| Autonomy policy | Settings for how the AI teammate acts in different scenarios: ask first, do then tell, or do silently. The person can change them, and every past approval and reversal is on record. | Examples 1, 2, and 3 | Prepare the repeated item in advance so approving it takes one click, and say the person still approves each one. Do not claim to act on its own, or to keep a grant the product cannot store. |

Every example needs at least one of these. **If you build only one example, build Example 1**, which needs only Autonomy policy.

## 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 a small orange square precedes every line the AI teammate speaks, so it is always clear who is talking, and the AI teammate's side sits on a warm, slowly moving gradient while the tool's side sits on flat gray. On the table, the AI teammate's note is a tinted row directly under the item it is about. A grant is a switch or a three-step control, left where it was until the person moves it; a switch turns on with a color change and a sliding knob. Use whatever your system already uses for "the AI teammate is speaking", "this is an offer", and "this is now on".

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

One table card, centered on a warm, slowly moving gradient of pink, orange and blue. Four column headings in gray: "Invoice", "Amount", "Sent", "Status". The first row, on a light gray tint, reads "Acme, week 13", "$4,200", "Today", with a "Needs you" tag. Directly under it, a tinted note row: the orange square and "I noticed something", then in gray "You’ve approved these invoices for 3 weeks straight with no changes. Want me to send them automatically?", and on the right a white toggle reading "Send these for me", its switch off. Below the note, two more rows: "Acme, week 12", "$4,200", "1 wk ago", "Approved", and "Acme, week 11", "$4,200", "2 wk ago", "Approved".

The switch is off at rest.

The loop starts on the table with the note's space empty and the rows below already in their places. The week 13 row takes its gray tint as a small rounded pill holding the orange square rises in the empty space, widens across the row, and "I noticed something" arrives one word at a time. The note opens downward to its full height, the gray line fades in, then the "Send these for me" toggle. The rows below never move. As the toggle arrives, a cursor fades in at the lower right; once it is in, it glides to the switch; the toggle takes a gray hover highlight and presses in slightly on the click. The switch turns green and its knob slides to the right. "I noticed something" and the gray line fade out, the orange square replays its entrance, and "Got it. I saved it to your preferences." arrives word by word in the space both lines took, with "preferences" underlined as a link. The week 13 row still reads "Needs you". The end state holds for about 3.5 seconds, then the loop restarts.

What to take from it: the offer comes from the history under it, three weeks of approvals with no changes, and says so (EA-R1). It is a question with a switch left off, so nothing changes until the person turns it on (EA-R2). Once the person agrees, the AI teammate says where the grant was saved, and the word "preferences" leads there (EA-R7). The whole exchange is one line and one control (EA-R8). On screen the week 13 row still reads "Needs you" after the grant; in a build the grant takes effect at once, so week 13 leaves the person's queue and is reported after it is sent (EA-R5).

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

The same card size and position, on a flat light gray background with no gradient. Square corners and a plain white card. No orange square, so nothing on screen speaks as an AI teammate. The same four column headings, then six rows: "Acme, week 13", "$4,200", "Today", "Needs you"; then "Acme, week 12" through "Acme, week 8", each "$4,200", sent "1 wk ago" through "5 wk ago", each "Approved".

What to take from it: week 13 waits on the person exactly as week 1 did, though five approvals with no changes sit right beneath it and nothing notices them (EA-R1). However long the record grows, it never changes how the next invoice is handled (EA-R5). 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 all three. 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. Want me to handle these? (Spot the pattern, ask to own it)

After the same report is approved a dozen times, a colleague asks “want me to just send these?”

- **Cue:** The person has approved the same action several times without changes
- **Move:** Points out the pattern and offers to handle it automatically
- **Response:** Hands it over, sets conditions, or declines
- **Resolution:** Repetitive approvals disappear, by agreement
- **Needs:** Autonomy policy.
- **Reference frame 05:** the orange square and the line "You’ve approved 12 Acme invoices unchanged." Under it, three gray rows, "Acme #1041", "Acme #1042", "Acme #1043", each marked "Approved" on the right. Under a thin divider, the row "Handle Acme invoices for me" with a switch on the right. In the animation the square swells, the line arrives word by word, the three invoice rows rise in one after another, and then the switch row rises in with the switch off. A cursor moves to the switch and clicks it; the switch turns black and its knob slides right. The end state holds for about 3.5 seconds, then the loop restarts. A loop of about 9.5 seconds. The offer names the run that earned it and shows a few of the approvals behind it, and the handover happens only when the person flips the switch.
- **Sample lines:** "You’ve approved 12 Acme invoices with no changes. Want me to send these for you?" / "Got it. I’ll send them and tell you after each one." / "No problem, I’ll keep bringing them to you."
- **Satisfies:** EA-R1, EA-R2, EA-R5, EA-R8.

### 2. I’ll take care of it (Widen the mandate gradually)

A colleague who has handled a task cleanly for a while asks to take on the adjacent work too.

- **Cue:** A run of actions in one area goes without correction
- **Move:** Offers to take on adjacent work or skip confirmations in that area
- **Response:** Widens the mandate, or keeps the current level
- **Resolution:** Autonomy grows in step with demonstrated reliability
- **Needs:** Autonomy policy.
- **Reference frame 06:** the orange square and the line "I can file the receipts that come with them, too." Under it, a small gray label "Receipts" over a three-part control: "Off", "Ask first", "Tell me after". In the animation the square swells, the line arrives word by word, and the control rises in with "Ask first" selected, shown as a dark thumb behind its label. A cursor moves to "Tell me after" and clicks it; the dark thumb slides from "Ask first" to "Tell me after", and the labels swap colors as it passes. The end state holds for about 3.5 seconds, then the loop restarts. A loop of about 9 seconds. The AI teammate asks for one adjacent task, not a broader mandate, and the person picks how far it goes from three levels instead of a yes or a no.
- **Sample lines:** "I’ve filed every Acme invoice cleanly for a month. I can file the receipts that come with them, too." / "Should I ask first on receipts, or just tell you after?" / "Okay, I’ll keep asking on receipts for now."
- **Satisfies:** EA-R3, EA-R4.

### 3. Oops, I’ll ask next time (Step back after a correction)

When something they did gets reversed, a colleague goes back to checking first on that kind of thing, without a fuss.

- **Cue:** The person reverts or overrides an AI teammate action
- **Move:** Returns to asking first in that area, and says so briefly
- **Response:** Leaves it there, or restores the autonomy when ready.
- **Resolution:** Corrections are cheap and never argued
- **Needs:** Autonomy policy.
- **Reference frame 07:** a light red banner with a small "!" icon and "You undid my change to the Birch invoice". Under it, the orange square and the line "I’ll ask before I touch invoices again." Then two buttons of equal width, "Let me act again" and "Ask me first". In the animation the banner rises in first, then the square arrives and swells and the line comes in word by word, and both buttons rise in, both light gray. A cursor moves to "Ask me first" and presses it, and it turns black. The end state holds for about 3.5 seconds, then the loop restarts. A loop of about 10 seconds. The AI teammate states the correction plainly, steps back in one line without defending what it did, and keeps the way to restore its autonomy one click away.
- **Sample lines:** "You undid my change to the Birch invoice. I’ll ask before I touch invoices again." / "Okay, I’ll go back to sending them without asking."
- **Satisfies:** EA-R6, EA-R7.

## Edge cases

- **No track record yet.** Make no offer; keep asking on each item until a run of unchanged approvals exists (EA-R1).
- **The person declines.** Keep asking as before, and do not offer again until the record has grown (EA-R8).
- **The task itself changes.** When an item no longer matches what was granted (a new amount, a new recipient), bring that one back to the person as outside the grant (EA-R5).
- **A correction after handover.** Go back to asking first in that area, and say so in one line (EA-R6).
- **The person restores autonomy.** Take it back up at the level the person picks, and say where it is saved (EA-R4, EA-R7).
- **The person revokes it later.** Stop at once, and put the items back in the person's queue (EA-R5, EA-R7).

## 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, hold the combined offers to EA-R8, so the person is asked to hand over one task at a time, never several at once.

- [Hand off judgment calls](https://humorphism.com/patterns/hand-off-judgment-calls.md): the decisions that stay with the person however long the record grows.
- [Learn from feedback](https://humorphism.com/patterns/learn-from-feedback.md): how corrections change what the AI teammate does next.
- [Remember across sessions](https://humorphism.com/patterns/remember-across-sessions.md): how a grant and its record carry from one session to the next.

## Checklist

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

- [ ] EA-R1: An offer to take on more appears only after a run of unchanged approvals, and states that run ("You’ve approved these invoices for 3 weeks straight with no changes").
- [ ] EA-R2: The offer arrives as a control left off, and nothing changes until the person turns it on.
- [ ] EA-R3: Each offer covers a single task or area, never a judgment call, such as a hiring or refund decision, or a decision with legal, safety, or reputational stakes, and comes only after a clean run in that task or in the area next to it.
- [ ] EA-R4: Each area shows its current level, the person can move it to any level directly, and the product's own levels map onto these three, with no silent level, since EA-R5 requires a report.
- [ ] EA-R5: Once granted, those items stop waiting on the person, the AI teammate handles them, and it reports at the level the person chose; an item outside what was granted comes back to the person.
- [ ] EA-R6: When the person reverts or overrides an action, that area goes back to asking first, and one line says so ("I’ll ask before I touch invoices again."), with a way to restore it.
- [ ] EA-R7: Each grant is saved where the person can see it ("I saved it to your preferences"), and turning it off works at once.
- [ ] EA-R8: The offer is a single line with a single control, and a declined offer does not return until the record has grown.
- [ ] 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/earn-autonomy), 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-notices-the-run.png](https://humorphism.com/patterns/earn-autonomy/01-teammate-notices-the-run.png) | The note under week 13, "I noticed something", with "Send these for me" off | AI teammate version |
| [02-teammate-person-reaches-the-switch.png](https://humorphism.com/patterns/earn-autonomy/02-teammate-person-reaches-the-switch.png) | The cursor on the "Send these for me" switch, the toggle highlighted | AI teammate version |
| [03-teammate-saved-to-preferences.png](https://humorphism.com/patterns/earn-autonomy/03-teammate-saved-to-preferences.png) | The switch on, "Got it. I saved it to your preferences." | AI teammate version |
| [04-tool-week-13-waits-like-week-1.png](https://humorphism.com/patterns/earn-autonomy/04-tool-week-13-waits-like-week-1.png) | Week 13 waiting on the person above five unchanged approvals | Tool version |
| [05-example-want-me-to-handle-these.png](https://humorphism.com/patterns/earn-autonomy/05-example-want-me-to-handle-these.png) | Twelve unchanged Acme approvals named, three listed, and the handover switched on | Example 1 |
| [06-example-ill-take-care-of-it.png](https://humorphism.com/patterns/earn-autonomy/06-example-ill-take-care-of-it.png) | Receipts moved from "Ask first" to "Tell me after" | Example 2 |
| [07-example-oops-ill-ask-next-time.png](https://humorphism.com/patterns/earn-autonomy/07-example-oops-ill-ask-next-time.png) | The undone Birch change, "I’ll ask before I touch invoices again.", and "Ask me first" pressed | Example 3 |
