MMarketing Against The Grain
← All frameworks
Innovation

Bespoke Interface Reasoning Ladder

Reason from user intent to a purpose-built interface

Difficulty
Advanced
Time to result
~months to results
Steps
6
Confidence
98%

The bespoke interface reasoning ladder converts an ambiguous natural-language request into a purpose-built application or result view. The system first determines whether the request needs a user interface at all; a concise text answer may be better for a simple question. If a richer experience is warranted, it evaluates whether it has enough information to proceed and identifies ambiguities that would materially change the output. It then asks focused clarifying questions, chooses an appropriate information architecture, and progressively reasons toward code and data. The generated result can be visual, interactive, and unique to the individual query rather than selected from a fixed template. This shifts AI products from answering inside a chat panel to assembling temporary software experiences around each user’s objective.

Origin

Extracted from Marketing Against The Grain

Core principles

  • 01Choose an interface only when the task needs one
  • 02Resolve consequential ambiguity before generating
  • 03Move from broad decisions to detailed implementation
  • 04Generate the experience around the user’s actual task
  • 05Use code and data to make the result interactive

How to run it

  1. 1

    Interpret the outcome

    Determine what the user is trying to accomplish rather than responding only to the literal words in the prompt.

    Pro tip Restate the likely outcome internally before choosing a response format.

    Watch out Over-inference can substitute the system’s preferences for the user’s intent.

  2. 2

    Choose whether to build a UI

    Decide whether plain text is sufficient or whether the request needs organized, interactive information.

    Pro tip Use a UI when users must compare, explore, configure, or act on several elements.

    Watch out Generating an application for every request creates unnecessary delay and complexity.

  3. 3

    Measure ambiguity

    Identify missing details that would substantially alter the content, structure, or recommendations.

    Pro tip Distinguish consequential ambiguity from details that can safely receive defaults.

    Watch out Proceeding through important ambiguity can produce a polished solution to the wrong problem.

  4. 4

    Clarify selectively

    Ask one or more focused questions that resolve the highest-impact uncertainties.

    Pro tip Offer constrained choices when users may not know how to specify their preferences.

    Watch out An exhaustive interview can become more burdensome than the original task.

  5. 5

    Design the information experience

    Choose the components, visual hierarchy, interaction model, and data needed to serve the clarified outcome.

    Pro tip Make every component answer a specific part of the user’s task.

    Watch out A visually rich interface can still be poorly matched to the decision being made.

  6. 6

    Generate and refine

    Produce the code and data, render the interface, and allow conversational iteration over the result.

    Pro tip Keep generated code editable so the experience can evolve without restarting.

    Watch out Generated code requires security, accessibility, and correctness checks before production use.

In the wild

Planning a child’s outdoor party

A parent asks for party ideas without specifying the child’s interests or the party format. The AI recognizes that the request needs organized options, asks a clarifying question, learns that the party will be outdoors and the child loves animals, and generates a visual interface for exploring relevant ideas.

The user receives a tailored, interactive planning experience instead of a generic list of links.

Choosing a software stack

A founder asks which tools to use for a new product. The AI determines that the answer depends on budget, traffic, team skills, and compliance needs, clarifies those constraints, then generates a comparison interface with costs, trade-offs, and an implementation path.

The founder can compare a personalized set of options and adjust assumptions interactively.

Common mistakes

Building before clarifying

Generating the interface immediately can lock an incorrect assumption into every downstream component.

Using one fixed result layout

A universal template defeats the purpose of matching the experience to the task.

Confusing visual richness with utility

The interface should improve understanding or action, not merely demonstrate generative capability.

Is it for you?

Best for

Complex requests that require organized information, interaction, personalization, or comparison rather than a single textual answer.

Not ideal for

Simple factual questions where generating an interface adds friction without improving comprehension or actionability.

From the transcript

Gemini uses a series of reasoning steps, going from broad decisions to increasingly high resolution of reasoning, finally getting to code and data.

Google AI demonstration narrator · 13:30

First, Gemini considers: does it even need an UI? Is a text prompt best?

Google AI demonstration narrator · 14:00

So it had asked a clarifying question.

Google AI demonstration narrator · 14:00

From the episode

Google Launches Gemini AI (And It’s Better Than GPT-4)