LLenny's Podcast
← All frameworks
InnovationKevin Weil (CPO at OpenAI, ex-Instagram, Twitter)

Reason About It Like a Human

To design or debug AI behavior, ask what an equivalent human would do in the same situation.

Difficulty
Easy
Time to result
~days to results
Steps
3
Confidence
85%

Because interacting with a model is so much like interacting with a person, you can borrow intuition from human behavior to design AI experiences and solve AI problems. When you're unsure how an AI product should work — or why some AI behavior is true — ask what an equivalent human would do in that situation and use it as a starting perspective.

Origin

Kevin Weil (CPO, OpenAI) cites this as the most counterintuitive thing he learned building AI products, drawn from designing OpenAI's reasoning-model UX and multi-model ensembles.

Core principles

  • 01Talking to a model is enough like talking to a human that human behavior is useful prior art
  • 02Use the human analogy to gain intuition and a different perspective — not always as the literal answer
  • 03The analogy scales from an individual human to a group of humans (e.g. brainstorming, a company)

How to run it

  1. 1

    Frame the AI problem as a human situation

    Restate the design or behavior question as: what would an equivalent human do here?

    Watch out This yields intuition and perspective, not always the final answer — don't apply it literally every time.

  2. 2

    Derive the design from the human answer

    Translate how a human would behave into the product's actual behavior or architecture.

  3. 3

    Scale the analogy from individual to group

    For harder problems, ask what a group of humans would do — brainstorming, dividing labor, integrating outputs — and mirror that structure with multiple models.

    Pro tip Better thinking often comes from several models attacking a problem and one model integrating their outputs, just like a brainstorm.

In the wild

Designing the reasoning-model 'thinking' UI

Facing the novel problem of a model that needs to sit and think for ~20 seconds, the team asked what a human would do: not go silent, but also not babble every thought. A person would say 'huh, good question' and give little updates. That human behavior became the shipped UI — light summarized updates rather than a silent spinner or the full raw chain of thought.

A consumer-friendly reasoning experience that felt natural rather than broken or overwhelming.

Common mistakes

Applying the human analogy too literally

The answer is sometimes not what a human would do; use the analogy to gain intuition, not as a rigid rule.

Is it for you?

Best for

AI product designers and builders stuck on how an AI experience should behave or why a model behaves as it does

Not ideal for

Low-level performance or systems problems where human intuition has no useful analogue

From the transcript

you can often reason about it the way you would reason about another human and it kind of works

36:00

sometimes to gain intuition for how you might solve a problem, you think about what an equivalent human would do in those situations

1:04:00

there's just like all these situations where you can actually kind of reason about it like a group of humans or an individual human and…

39:00

From the episode

OpenAI’s CPO on how AI changes must-have skills, moats, coding, startup playbooks, more

Kevin Weil (CPO at OpenAI, ex-Instagram, Twitter)