Skip to content
El Observatorio

Building & Agents · 2 min de lectura

Let Your Agent Ask the Wrong Questions, Lose Everything

Before you build an agent to serve a customer need, you have to be certain the need is real — and the same conversational traps that fool founders in human interviews will fool the agents you train on bad signal.

Con · estudiado y reformulado para builders AI-native“The Mom Test” — Rob Fitzpatrick ↗

The Mom Test Problem Scales With Automation

Rob Fitzpatrick's The Mom Test makes one hard point: people will lie to your face without meaning to, because compliments are socially easy and honest criticism is not. Ask someone if they'd use your product and they say yes. Ask them when they last paid to solve the problem and the answer tells you everything. For an AI-native company, this matters double. If you train workflows on customer feedback that was never honest, you bake the lie into the system permanently.

Past Behavior Is the Only Ground Truth

Fitzpatrick's discipline is to ask about life, not ideas — specific past actions, real costs, actual workarounds already in use. That same discipline applies when you design evaluation harnesses for agents. Your evals should reflect what customers actually did, not what they said they wanted. A red case drawn from a real failed interaction beats a hundred hypothetical test cases. Build the eval before you build the agent, and populate it with behavioral evidence, not stated preference.

Commitment Signals Tell You What to Automate

The Mom Test measures real interest through commitment: time given, money paid, access granted. Before you graduate a workflow to a full agent — letting it choose its own path rather than follow a fixed one — demand the same proof. Has a real user paid, waited, or changed behavior to get this outcome? If the answer is no, you have a workflow problem wearing an agent costume, and no amount of clever prompting will save it.

Human Judgment Holds the Standard

Even with honest signal and rigorous evals, someone has to decide what the agent is actually for. Fitzpatrick's framework assumes a founder who stays in the room, resists flattery, and keeps asking harder questions. In an autonomous system, that role does not disappear — it shifts. Humans set the goals, define the boundaries, and own accountability for what the agent does in production. The discipline of the Mom Test is, at bottom, the discipline of not outsourcing your judgment before you've earned the right to.

Prueba el Desktop OS

Explora la escena startup de Atlanta, en vivo.

Todo lo que acabas de leer vive dentro de nuestro Desktop OS — un espacio de trabajo interactivo que se abre directamente en tu navegador. Su mesa de Atlanta mapea la escena startup de la ciudad: quién está construyendo, qué está pasando y dónde. Ábrelo y échale un vistazo.

Abre el Desktop OS →