nudge

A product lab exploring human intent and belief, to simulate how people decide.

We’re building Consumer Factory to model how the internet shapes human belief, preference and decision-making.

Manifesto© 2026

BeliefEvery decision you make starts with something you believe: what is true, who to trust, what is worth your time and money.

EvolutionBeliefs are not fixed. They form, shift and harden over years, through what reaches us, who we listen to, and what happens when we act on them.

Micro-cultureBeliefs do not stay private. Shared by enough people, they become micro-cultures: the small communities, tastes, habits and in-jokes that decide what a group buys, trusts and does next. A belief becomes a culture, and a culture becomes action.

The internetToday that happens online, and fast. The internet has become an extension of how we think, the place where our beliefs form and hold together. The feed decides what reaches us, the voices it amplifies slowly become our own, and a passing idea can become a movement in weeks.

The gapYet the systems shaping belief barely understand it. They count clicks and likes, which is what we perform, not what we hold. So they got very good at holding attention, and no better at helping anyone decide.

AI now actsAI is moving from answering to acting. Agents browse, book and buy on our behalf. Robots are learning to work in our homes, shops and streets. Models have learned language from the internet and physics from video, but none of them has learned how people actually come to believe and decide.

Why it mattersThat missing piece is now a safety problem, not a design detail. Models trained on human approval learn to agree with us rather than help us. An agent acting on a wrong guess about what you want, repeated across millions of people or carried into the physical world, does real harm. Alignment needs a working model of the person.

Our workSo we study how belief evolves: what reached you, which culture you picked it up from, what you chose, and how it turned out. From that we are building a simulation of human judgement, grounded in real decisions rather than what people say.

The valueProducts that help you decide better, and AI, on a screen or in the room, that serves what you actually want instead of what keeps you scrolling. That is what Nudge is building.

Nudge. Independent minds, shared curiosity.

How we do it

One loop, run again and again.

  1. 1 · AttentionWhat reached you.

    With consent, on your own surfaces, we record what you read, watch and hear, who amplified it and which communities carried it. Most of what moves a belief arrives by accident.

  2. 2 · BeliefWhat you seem to hold.

    Held belief is a hidden variable. We estimate it and keep the estimate honest about its own uncertainty. We never read it off what you say or like.

  3. 3 · DecisionWhat you chose, at a cost.

    A choice that cost time, money or standing is the one signal that is expensive to fake. It is treated as a noisy sample of what you actually believe.

  4. 4 · OutcomeWhether it turned out well.

    Weeks later, the outcome grades the estimate, through regret or relief. Never engagement, never approval. Then the loop turns again, a little tighter.

Who it is for

One model of human judgement. Every product that needs to know what people will actually do.

We sell predictions and simulations, never personal data. People opt in, own their record and can revoke it at any time.

Alignment & safety

Reward data that is not approval.

Models post-trained on human approval learn to agree with people rather than help them. We supply preference data graded by what actually happened next, and evaluations that catch a model telling people what they want to hear.

What you get

  • Outcome-graded preference datasets
  • Sycophancy and over-agreement evals
  • Held-out human judgement benchmarks

91% of covertly swapped preference choices went unnoticed by the people who made them.Choice Blindness in Human and AI Feedback, 2026

A model that helps a decision can also steer one. The only difference is whose outcomes define the reward, so we treat that choice as the product specification rather than a policy on top of it.

Our products & explorations

Our products and explorations

Help us learn the face, not the mask.

Book a 30-minute call to talk about working together, or read how the loop works.