01 ambition
Speed
The same work in less time, the bread and butter for AI's economic value. Well established and requires little creativity in how you approach AI.
> stance × structure · 14 stances wired to 7 components
The Matrix
Which Stack layer carries each Stance.
The Stances describe the contract a human thinks they have with an AI system: “I am asking an Oracle”, “I am handing work to an Executor”. The Stack describes what the human must actually supply for any contract to hold: instruction, information, purpose, doubt, consistency, measurement, and a place for the work to run.
Each stance leans on one layer more than the others. That layer is where the stance’s characteristic failure lives, and therefore where the guardrail belongs. Knowing which stance you are in tells you which layer to check first. That is mode fluency in one move.
§1 matrix.circuit
The full circuit
Hover, tap or tab to a stance and follow the current to its layer. Select a layer to see every stance that leans on it. Click a stance to print its guardrail in the system log. Arrow keys move along a column; right and left cross the wire.
stance
layer
Oracle —load→ Judgment
stance Oracle
expects Truth to a hard question
load JUDGMENT
fail Confident plausible prose taken as fact
guard Ask for the uncertainty before the answer. Treat the answer as a claim, not a verdict.
§2 matrix.table
The same matrix, as text
The full table from MODE_FLUENCY_MATRIX.md, in its row order. It carries everything the circuit shows, without motion and without a pointer.
| Stance | What you expect | Load-bearing layer | Characteristic failure | Guardrail (the cardinal rule) |
|---|---|---|---|---|
| Oracle | Truth to a hard question | Judgment | Confident plausible prose taken as fact | Ask for the uncertainty before the answer. Treat the answer as a claim, not a verdict. |
| Genie | A discrete deliverable | Prompt | Literal compliance with a brittle wish | Specify scope, boundaries and success criteria before you ask (an Intent Doc). |
| Interpreter | Decode material I supply | Evaluation | Confabulated meaning that sounds insightful | “Show me exactly which parts of the text support each conclusion.” Never accept an interpretation you cannot trace. |
| Executor | A delegated job, reviewed after | Harness + Judgment | Automation complacency; locally correct steps compounding into the wrong outcome, found late | Plan before action, checkpoints, a deviation report, nothing consequential ships uninspected. Delegation transfers labor, never responsibility. |
| Miner | The shape of a space, not one answer | Context (writing) | Believing an anomaly because you wanted it; losing the vein | Assay cold, later. Record every find when you find it: if you didn’t record it, you never had it. |
| Tutor | A path to the answer | Intent | Being taught toward a goal you never stated | State what you want to be able to do afterwards before the first question. |
| Adversary | Attack on my plan | Intent | Critique of the wrong thing | Name what you are defending and what would change your mind. |
| Scribe | Craft applied to my substance | Prompt | Polished form over a hollow draft | Provide the substance and the format; review for what was invented to fill gaps. |
| Simulator | A rehearsal of reality | Context | A rehearsal of a scenario that is not yours | Load the real constraints, the real counterpart, the real stakes. |
| Librarian | Synthesis of what is known | Evaluation | Confident retrieval of what was never said | Check the citation exists and says what is claimed. |
| Mirror | My own thinking, clarified | Coherence | Flattery mistaken for reflection | Ask for the structure of your argument back, not for agreement. |
| Translator | A bridge between two worlds | Context | A bridge from one world to a caricature of the other | Supply both maps; name both audiences. |
| Therapist / Confessor | A space to think aloud | Judgment | Dependency; outsourcing judgment about your own life | Decide in advance what you will not delegate. |
| Muse | Provocation, not correctness | (none) | There is no failure, only dross | None. This is why the Muse stays hidden: it cannot be metric’d. |
§3 how_to_use
How to use it
- Before you start: name the stance out loud. Then look up its layer and do the one thing in the guardrail column.
- When it goes wrong: do not rewrite the prompt first. Find the stance you were in, go to its layer, and re-enter there.
- For a team: the stances your team uses most tell you which layers your shared setup must make strong. A team that delegates (Executor) needs harness and judgment work; a team that researches (Librarian, Interpreter) needs evaluation.
§4 axis 03
Autonomy: how much of the work do I watch?
As autonomy increases, spec rigor must increase proportionally. In Assistive mode, your brain is the spec. In Autonomous mode, the written spec is the only control mechanism between intent and execution.
| Mode | How it works | Human understanding | Risk of incorrect output | Spec rigor required |
|---|---|---|---|---|
| Assistive | You type, AI suggests | Very High | Very Low | Low |
| Directed | You describe, AI types, you review | High | Low–Medium | Medium |
| Autonomous | You assign, AI executes, you review the result | Medium | Medium–High | High |
| Autonomous + Spec | Autonomous, with a written spec | Medium | Medium (reduced) | High (by definition) |
| Continuous | AI runs as a background service | Low | High | Very High |
Adding a spec is the only intervention in the table that improves a risk rating. The smallest spec is the four-line Intent Doc: TASK (what specifically was asked for), SCOPE (who or what is affected, and excluded), BOUNDARIES (what the AI will not do), IF UNCERTAIN (ask, flag, or stop).
As autonomy rises, the stances shift toward the Executor and the load shifts toward Harness, Judgment and Evaluation. This is why the AI @ Workplace curriculum spends 101–301 on chatbots and 401 onward on agents: the layer that carries the load moves as the human steps back. Productive people also switch modes on a single task; staying in one mode regardless of the task is its own anti-pattern (mode rigidity).
§5 axis 04
Ambition: how big is the ask?
adopts Felix Rieseberg's four types of ambition, from “How to be more ambitious: A tutorial for humans with models” (13 October 2026). The descriptions below are his.
02 ambition
Altitude
Going up in abstraction and considering the outcome rather than the individual task.
03 ambition
Depth
Considering a task the way a field-leading expert would. Rather than operating at the task's layer, we go one level deeper and touch the underlying data, tools, and problems.
04 ambition
Range
Drawing on multiple fields rather than just the one the task might need at the surface, using knowledge, tools, and techniques you normally wouldn't.
§6 lineage
Lineage
Stances build on Archetypes of LLM-Human Interaction (v3). Related framings classify by autonomy or social role rather than by the human's expectation: Anthropic's AI Fluency (4Ds: Delegation, Description, Discernment, Diligence), Mollick's centaurs and cyborgs, Microsoft's “agent boss”, and the CHI 2026 “human-LLM archetypes” paper for decision support (arXiv 2602.11924). The Stack's intent engineering layer is credited to Nate B. Jones.