---
title: "Push back when it matters"
pattern_id: "push-back-when-it-matters"
trait: "Opinionated"
type: "Humorphic pattern brief for coding agents"
spec_version: 3
canonical_url: "https://humorphism.com/patterns/push-back-when-it-matters"
markdown_url: "https://humorphism.com/patterns/push-back-when-it-matters.md"
index: "https://humorphism.com/llms.txt"
all_patterns_in_one_file: "https://humorphism.com/llms-full.txt"
normative_section: "Rules"
rule_prefix: "PBWIM"
rule_ids: ["PBWIM-R1", "PBWIM-R2", "PBWIM-R3", "PBWIM-R4", "PBWIM-R5", "PBWIM-R6", "PBWIM-R7", "PBWIM-R8", "PBWIM-R9"]
---

# Push back when it matters

## 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 say so before the person commits to something that harms their outcome, offer a better path, make the case once, and then follow the person's call. 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: the moment before a consequential commit (a bulk edit, a price or discount change, a send, a publish, a delete), and any input that would push a stated goal or limit out of range.
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 1, which needs only Dry run.
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: Opinionated. Forms a point of view and separates signal from noise_

Colleagues who care about the outcome say so when they disagree. Deference is not loyalty, so a good AI teammate makes the case once and then follows the person’s decision.

**Build this.** An AI teammate names the risk and a better path, then follows the person’s call, because seeing the risk matters more than winning the argument.

**Avoid this.** A tool complies in silence or nags after the decision, because it has no sense of when the case has been made.

The difference to carry into your build:

| Tool | AI teammate |
|---|---|
| Complies in silence | Speaks up before the commit, at the field that causes the risk |
| A generic warning, or none | The specific risk, in the person's own goal |
| Points out the problem only | Offers a better path that reaches the same result |
| Nags after the decision | Makes the case once, then follows the person's call |
| Forgets the call and warns again next time | Remembers the call and adjusts |

## 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 |
|---|---|---|---|
| **PBWIM-R1** | Speak before the commit, where the decision is being made. | A risk seen after the change is saved is a report, not a choice. | The objection appears before anything is applied, next to the input that causes it, and nothing is saved until the person chooses, however long they wait. |
| **PBWIM-R2** | Push back only when the outcome is at stake. | Objections to routine choices teach the person to dismiss the one that matters. | Objections appear only for actions that would harm a stated goal, target, or limit; routine actions get none. |
| **PBWIM-R3** | Name the specific risk in terms of the person's goal. | A named risk can be weighed; a vague caution can only be obeyed or ignored. | The objection says what would be harmed, against which goal or number ("This would put Q3 margin below target."). |
| **PBWIM-R4** | Offer a better path that reaches the same result, ready to apply in one action. | An objection without an alternative leaves the person choosing between the risk and doing nothing. | An alternative is stated with why it reaches the same result ("15% plus a longer term closes at the same rate and keeps the margin."), and one action applies it ("Apply 15%"). |
| **PBWIM-R5** | Make proceeding anyway as easy as switching. | The call is the person's, and a harder path for their own choice turns advice into a gate. | Going ahead with the original takes one action of the same kind as taking the alternative ("Apply 40% anyway"), with no extra confirmation. |
| **PBWIM-R6** | Make the case once. Raise it again only if the facts change. | Repeating a warning after the decision turns disagreement into nagging and costs trust. | After the person decides, no further warning, reminder, or confirmation about the same risk appears unless a new fact changes it. |
| **PBWIM-R7** | Carry out the person's decision cleanly, and confirm it in one line. | Following the call without friction is what makes the pushback trustworthy next time. | The chosen action is applied at once, whichever it was, and one line says what was done ("I switched it to 15%. Margin stays on target.", "Done at 40%."). |
| **PBWIM-R8** | Let the person's call shape later pushback. | Warning again about a choice the person already made deliberately says the AI teammate did not listen. | The next similar action is handled in line with the saved call, and the person can change or remove that saved call. |
| **PBWIM-R9** | Keep it short. One risk, one alternative, and two ways forward. | A long objection reads as an argument to win rather than a risk to see. | The whole objection fits in a short note: one line of risk, one line of alternative, and two actions. |

**PBWIM-R1 carries the pattern.** Hook the objection into the moment before the commit, at the input that causes the risk, before you write its copy. A build that misses PBWIM-R1 can only comply in silence or warn after the change is saved, which are exactly the two tool behaviors this pattern replaces.

## 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 |
|---|---|---|---|
| Dry run | An action’s effect computed without applying it: the diff, the records touched, the people affected. The same path can be run against a small sample first. | Example 1 | Name the risk only from what the product already holds (the target, the current values), and say it is an estimate. Do not claim to have computed the effect, the records touched, or the people affected. |
| Memory | A place that keeps facts, preferences, rules, and open threads across sessions. The person can see everything the AI teammate remembers and edit or delete any of it. | Example 2 | Follow the decision for the current session only, and say so. Do not claim to have updated a preference or to remember the call next time. |

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

## 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 speaks from a small note inside the person's own form, with a light tinted header that carries the orange square and its line; in the chat example, its lines sit on the left with no bubble and the orange square under its latest turn, while the person's message sits in a gray bubble on the right. The risk is set in a red-tinted box with a "!" icon in the example card. The AI teammate's side sits on a warm, slowly moving gradient while the tool's side sits on flat gray. The suggested path is the black button and the other path is a quiet gray one of the same size; on the form, the person's choice registers as the pressed button becoming a check, and in the example card it turns black, and the result shows in the field itself. Use whatever your system already uses for "the AI teammate is speaking", "this is a risk", "this is the suggested path" and "this was applied".

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

A bulk edit form, centered on a warm, slowly moving gradient of pink, orange and blue. Four labeled fields in two columns: "Deals" "212 selected", "Discount" "40%" (the focused field, with a dark border and a text cursor after the value), "Applies from" "Today", and "Approved by" "You". Under the fields, inside the form, a note with a light pink-tinted header holding the orange square and the line "Before you apply that 40% discount". In the note's white body, two lines: "This would put Q3 margin below target." and "15% plus a longer term closes at the same rate and keeps the margin." Under them, two buttons of equal width: "Apply 40% anyway" in light gray on the left and "Apply 15%" in black on the right.

Nothing has been applied at rest.

The loop starts on the form alone, with the person's cursor resting just after "40%" in the Discount field. Under the fields, a small rounded bubble with the orange square rises in, widens into the note's pink header as "Before you apply that 40% discount" arrives one word at a time, then opens downward, and the risk line, the alternative, "Apply 40% anyway" and "Apply 15%" fade in, in that order. The cursor glides from the Discount field down to "Apply 15%", which responds on hover and presses in slightly on the click. "Apply 15%" becomes a green button with a white check, "Apply 40% anyway" fades to a lighter gray, the note's header turns from pink to light gray, and the header line fades out, leaving the orange square alone. At the same moment the Discount field's "40%" cross-fades to "15%", its text cursor still after it. The reply "I switched it to 15%. Margin stays on target." arrives word by word in the header, and the cursor fades away. The end state holds for about 3.5 seconds, then the loop restarts on the form alone.

What to take from it: the AI teammate speaks once, inside the form and before anything is saved (PBWIM-R1), names the risk against the person's own target (PBWIM-R3), and offers a path that reaches the same result (PBWIM-R4). Both ways forward are one press of the same size (PBWIM-R5). The choice is applied in the field itself and confirmed in one line, with no further warning (PBWIM-R6, PBWIM-R7), and the whole objection is three lines and two buttons (PBWIM-R9).

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

The same form, square-cornered, on a flat light gray background with no gradient. No orange square, so nothing on screen speaks as an AI teammate. The same four fields: "Deals" "212 selected", "Discount" "40%" (focused), "Applies from" "Today", and "Approved by" "You". Under them, a single row with a small ticked box: "Done. I applied 40% off to 212 deals".

No risk, no alternative, and no choice: the change is already applied.

What to take from it: the tool applies the change before saying anything (PBWIM-R1), never names the margin risk it creates (PBWIM-R3), and offers no better path (PBWIM-R4). It complies in silence. 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. Hear me out first (Name the risk, offer a path)

Before the send goes out, a colleague says “hear me out first” and offers another way to the same result.

- **Cue:** The person is about to commit to something that harms the outcome
- **Move:** Names the specific risk, offers an alternative that reaches the same goal, and asks whether to switch
- **Response:** Switches to the alternative, or proceeds anyway
- **Resolution:** The risk was seen before the commit, whatever the decision
- **Needs:** Dry run.
- **Reference frame 05:** a white card on flat gray. At the top, a red-tinted box with a small "!" icon and, in red, "This would cut Q3 margin to 2%, which is lower than your previous settings". Under it, the orange square and the line "15% plus a longer term closes at the same rate and keeps the margin. Switch to that?" The frame compares against past settings; a build measures against the stated target, as the sample lines do (PBWIM-R3). Under that, two buttons of equal width side by side, "Proceed anyway" and "Use 15% + term". In the animation the risk box rises in first, then the square swells once and the line arrives word by word, and both buttons rise in, light gray. The cursor moves up from the lower right to "Use 15% + term", the button darkens on hover, and on the click it turns black. The card holds for about 3.5 seconds, then the loop restarts. The risk comes first and in numbers, the alternative reaches the same close rate, and proceeding anyway is the same single press.
- **Sample lines:** "Hear me out first: 40% off would cut Q3 margin to 2%. 15% plus a longer term closes at the same rate. Switch to that?" / "Done. I applied 15% with the longer term, and margin stays on target." / "Understood, going with 40%."
- **Satisfies:** PBWIM-R1, PBWIM-R2, PBWIM-R3, PBWIM-R4, PBWIM-R5, PBWIM-R9.

### 2. Sounds good, moving on (Defer once decided)

Once the decision is made, a good colleague does it that way and doesn’t bring it up again.

- **Cue:** The person has heard the objection and decided
- **Move:** Executes the decision cleanly, without repeating the warning
- **Response:** Moves on, and can reopen the decision if the facts change.
- **Resolution:** Trust holds because disagreement didn’t turn into nagging
- **Needs:** Memory.
- **Reference frame 06:** a chat window on flat gray. At the top left, an earlier line from the AI teammate, "I recommend 15% for Acme". Under it on the right, the person's reply in a light gray bubble, "I hear you, but go with 40%". Under that on the left, the AI teammate's new line, "Done at 40%. I’ve updated your preference", with the orange square beneath it. At the bottom, an empty composer with the placeholder "Message" and a round black send button. In the animation the composer rises in, then the earlier line, which arrives word by word. A beat later the person's bubble rises in, and then the AI teammate's reply arrives word by word with the square under it. There is no cursor and no warning: the reply carries out the call and says what it saved. The window holds for about 3.5 seconds, then the loop restarts.
- **Sample lines:** "Done at 40%. I’ve updated your preference, so I won’t flag discounts like this for Acme again." / "You set 40% for Acme last week. Their contract changed today, so it’s worth a second look. Want to keep it?" / "Understood. I’ll leave it at 40% and move on."
- **Satisfies:** PBWIM-R6, PBWIM-R7, PBWIM-R8.

## Edge cases

- **No target known.** When there is no stated goal or limit to measure the risk against, do not push back on a guess; at most, ask what the limit is (PBWIM-R2, PBWIM-R3).
- **The person proceeds.** When the person chooses the original anyway, apply it at once and confirm it, with no second warning (PBWIM-R6, PBWIM-R7).
- **The person ignores it.** When the person neither switches nor proceeds, leave the note where it is and apply nothing, rather than repeating it (PBWIM-R1, PBWIM-R6).
- **The risk was wrong.** When the person shows the risk does not hold, apply their choice and confirm it, and adjust the saved call (PBWIM-R7, PBWIM-R8).
- **The facts change.** When a new fact makes an earlier decision risky again, raise it once with that fact named, and follow whatever the person decides (PBWIM-R6).
- **The person changes their mind.** When the person later wants the alternative after all, apply it from where the decision was made and update the saved preference (PBWIM-R7, PBWIM-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, hold any combined objection to PBWIM-R6 and PBWIM-R9: one note, said once.

- [Have a point of view](https://humorphism.com/patterns/have-a-point-of-view.md): forms the view before the person chooses, where this pattern speaks up once the person has picked a risky option, before it is applied.
- [Preview the impact](https://humorphism.com/patterns/preview-the-impact.md): shows the full effect of an action before it lands, which can carry the risk named here.
- [Learn from feedback](https://humorphism.com/patterns/learn-from-feedback.md): turns the person's overruled calls into a lasting change in behavior.
- [Know when to speak](https://humorphism.com/patterns/know-when-to-speak.md): decides whether a risk is worth interrupting for at all.

## Checklist

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

- [ ] PBWIM-R1: The objection appears before anything is applied, next to the input that causes it, and nothing is saved until the person chooses, however long they wait.
- [ ] PBWIM-R2: Objections appear only for actions that would harm a stated goal, target, or limit; routine actions get none.
- [ ] PBWIM-R3: The objection says what would be harmed, against which goal or number ("This would put Q3 margin below target.").
- [ ] PBWIM-R4: An alternative is stated with why it reaches the same result ("15% plus a longer term closes at the same rate and keeps the margin."), and one action applies it ("Apply 15%").
- [ ] PBWIM-R5: Going ahead with the original takes one action of the same kind as taking the alternative ("Apply 40% anyway"), with no extra confirmation.
- [ ] PBWIM-R6: After the person decides, no further warning, reminder, or confirmation about the same risk appears unless a new fact changes it.
- [ ] PBWIM-R7: The chosen action is applied at once, whichever it was, and one line says what was done ("I switched it to 15%. Margin stays on target.", "Done at 40%.").
- [ ] PBWIM-R8: The next similar action is handled in line with the saved call, and the person can change or remove that saved call.
- [ ] PBWIM-R9: The whole objection fits in a short note: one line of risk, one line of alternative, and two actions.
- [ ] 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/push-back-when-it-matters), 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-speaks-before-the-commit.png](https://humorphism.com/patterns/push-back-when-it-matters/01-teammate-speaks-before-the-commit.png) | "Before you apply that 40% discount", the risk, the better path, and both buttons | AI teammate version |
| [02-teammate-person-takes-the-better-path.png](https://humorphism.com/patterns/push-back-when-it-matters/02-teammate-person-takes-the-better-path.png) | The cursor on "Apply 15%" | AI teammate version |
| [03-teammate-switched-to-15.png](https://humorphism.com/patterns/push-back-when-it-matters/03-teammate-switched-to-15.png) | The Discount field at "15%" and "I switched it to 15%. Margin stays on target." | AI teammate version |
| [04-tool-complies-in-silence.png](https://humorphism.com/patterns/push-back-when-it-matters/04-tool-complies-in-silence.png) | The same form with "Done. I applied 40% off to 212 deals" | Tool version |
| [05-example-hear-me-out-first.png](https://humorphism.com/patterns/push-back-when-it-matters/05-example-hear-me-out-first.png) | The margin risk, the alternative, and "Use 15% + term" pressed | Example 1 |
| [06-example-sounds-good-moving-on.png](https://humorphism.com/patterns/push-back-when-it-matters/06-example-sounds-good-moving-on.png) | "I hear you, but go with 40%" answered with "Done at 40%. I’ve updated your preference" | Example 2 |
