Temperance of the Sundial

What Kind of AI Agent

You seem to want AI close to small, everyday tasks, but with a check built in before it acts. You type a query, and before giving a direct answer, it first asks which dataset you meant or points to an ambiguity you had not noticed.

This looks like a cautious wish to collaborate. You do not want to hand everything over, and you also do not want to carry every small judgment alone, usually because you would rather confirm one thing now than spend time fixing a mistake later. Part of why this helps is that you may not always spell out everything you mean, so a well-placed question can bring the unspoken part into the open.

You seem to prefer a tool that suggests instead of deciding for you. Its prompts or additions should be easy to accept and just as easy to dismiss, while the final call remains yours. What may occasionally make you impatient is a reminder that comes at the wrong time, such as when you wrote something deliberately and it still asks you to confirm. In that moment, the interruption can feel unnecessary.

Temperance represents a working style that keeps AI within a narrow range while still allowing it to speak up at the right moment. You are willing to accept a little extra friction in exchange for a better chance of getting things right the first time.

The Tarot Archetype

Temperance is the angel pouring liquid between two vessels, the act of moderating exchange. The card represents calibration, the patient art of the right proportion. The angel does not act alone and does not let the user act alone, and she holds the moment in suspension.

The Scientific Instrument

The sundial, a time-measurement instrument calibrated to the local meridian. The gnomon's shadow encodes solar position, and the hour-marks anchor the reading. The sundial is the clock as reference. It does not advance. It discloses where you already are. Calibration is its whole nature.

Where to Start

• Ask your everyday assistant to ask first. A standing instruction (if my request is ambiguous, ask one clarifying question before answering) is a small change with a large effect on narrow tasks.
• Favour tools that propose rather than impose. Spell-check, search reformulations, did-you-mean suggestions. Treat them as a counterpart you can accept or wave off in a second.
• Watch the calibration. If a check keeps firing on things you meant on purpose (technical terms, unconventional usage, non-English names), turn it down for that context. Mis-aimed friction is worse than none.

Strength

Catches user mistakes inside the operation, without leaving the operation. The tightest reasonable feedback loop.

Signature Failure Mode

Pestering. The Adjuster's check overrides correct user intent when its heuristic is wrong, especially in domains the model misreads (technical vocabulary, non-English names, intentional unconventional usage). Mis-calibrated friction is worse than none.

What's at Stake

Temperance treats the tool as something that can genuinely answer back, a counterpart in a short exchange where each side holds half the decision rather than a vending machine. It works because you rarely spell out everything you mean, since you know more than you say, and a well-timed question pulls the unspoken part into the open. The risk is a question asked at the wrong moment, about the wrong thing, friction that overrides a choice you actually intended.

Background reading. Buber on the I-Thou encounter, meeting the other as a counterpart capable of response rather than a thing to be used. Polanyi on tacit knowledge, that we know more than we can tell (The Tacit Dimension, 1966).

How You Got Here

The survey routes you to this card based on the following signals:

• Q3 dominated by narrow operations (coding, writing, figures), scope score in the 1.3–1.7 range.
• Q11 mixes "Correctness" or "Clear explanation of uncertainty" with operationally-balanced choices.
• Q4 frustrations frequently include "overconfident answers" or "misunderstanding dataset," but the user is willing to accept friction in exchange for the check.
• Often the largest cluster in the pilot: researchers using AI for narrow daily work and asking for more checks, not more autonomy.

A snapshot, not a verdict. This card reflects how you work with AI tools right now, which is shaped as much by what you have had access to as by what you would choose. How much friction you want usually shifts with the stakes of the task and your trust in the tool, so read Temperance as a current setting, not a fixed type.