The Wheel of the Orrery

What Kind of AI Agent

You open your inbox and the report is already there. The eval suite re-ran overnight on the new data, and a flagged line tells you metric three has moved because the system was already set to watch for that change.

You seem comfortable setting fairly broad pieces of work in motion and letting them run on their own rhythm, a validation that triggers when new data arrives, a weekly paper sweep that finishes on schedule, all on the system's timing and without you having to start each run yourself.

You do not seem especially attached to being present for every step, since you would rather encode a decision once and let it keep working. That reflects a strong ability to use tools and process as part of the work itself. You treat schedules, automation, and systems as part of how thinking gets done, letting them create steady forward motion even while your attention is elsewhere.

What frustrates you most seems to be operational failure, such as compute getting stuck, dependencies breaking, or context being lost. As for the system acting on its own, you mostly seem to welcome it. That is why you likely care about the things that keep unattended cycles trustworthy, such as reproducibility, experiment tracking, and outputs landing somewhere you will actually see them.

The risk is that automated cycles can drift quietly. They keep running on the inputs they were given, and when the data or code changes underneath them, they may carry the wrong version all the way to a polished but incorrect result. Wheel represents a working style that turns repeated decisions into automatic mechanisms. Its value is that it reduces the burden of starting the same work again and again, as long as you still check occasionally that it is doing what you originally meant.

The Tarot Archetype

The Wheel of Fortune turns by itself, and we ride. The card is about cyclical, system-driven action, the older fortuna that means the patterned recurrence of work rather than random fortune in the modern sense. Things happen on the Wheel's schedule rather than yours.

The Scientific Instrument

The orrery, a mechanical model of the solar system, driven by clockwork, running continuously and cyclically. The orrery is the canonical instrument of scheduled astronomy, used in 18th-century education and prediction. Its gears do not measure once, they keep measuring.

Where to Start

• Put one forgettable task on a schedule. A weekly data-quality sweep, a re-run of the eval set, a recurring reproducibility check, something broad you keep meaning to do and not doing.
• Make the output land where you will see it. A summary comment on the commit, a message in a channel you read, a short report in your inbox. An unread report is the same as no report.
• Pin the inputs. A cycle assumes the data and code it started with. When those change quietly, the job runs the wrong phase to completion. Schedule a check that it is still doing what you meant.

Strength

Ensures broad subtasks happen even when you forget. Time-shifted work is the Wheel's specialty.

Signature Failure Mode

Takes broad action you didn't sanction. If the phase scope drifts, such as new data or new code, the Conductor runs the wrong phase to completion before you notice. Cycles assume their inputs, and when inputs change quietly, the output is silently wrong.

What's at Stake

The Wheel is the part of your working system that remembers and acts for you. Your research mind does not stop at your skull. It extends into the tools, schedules, and pipelines around you, and the Wheel is the piece that keeps a cycle turning while your attention is elsewhere. The risk is drift. A cycle keeps running on the inputs it was handed, and when the data or the code quietly changes underneath it, it will carry the wrong phase all the way to a confident, finished, wrong report before anyone looks.

Background reading. Hutchins on distributed cognition, the cognitive system extending beyond any single mind (Cognition in the Wild, 1995). Wiener on the cybernetic feedback loop, work as a controlled cycle with a human supervisor.

How You Got Here

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

• Q3 leans broad: literature search, EDA, brainstorming, model implementation.
• Q11 emphasizes "Reproducibility," "Experiment tracking," or "Dataset understanding," pulling initiative toward the system.
• Q10 (setup-burden willingness) reads heavy on "lowers cognitive load," "integrates literature/code/data," and "catches mistakes": strong system-init demand.
• Q4 frustrations are operational rather than oversight-driven, such as compute, dependencies, or losing context; the respondent welcomes the system taking initiative.

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. Running work on the system's own schedule usually follows from having the infrastructure set up to do it, so the Wheel reflects your current tooling as much as your preference.