PATTERN 16 March 2026

The Four Gears

Your AI thinks in one gear. It needs four.

The Four Gears

You already switch thinking modes dozens of times a day. When you review a colleague’s proposal, you are evaluating. When you brainstorm with your team, you are generating. When you cross-reference a finance model with a biology paper, you are engaging across domains.

You know these modes exist. You have never named them.

The Problem: Tasks Without Posture

Every major cognitive framework stops at what you do. Bloom’s Taxonomy gives you six cognitive levels (remember, understand, apply, analyse, evaluate, create). De Bono gives you six hats. AI agent architectures give you eight verbs, twelve tools, forty-seven functions.

None of them answer a simpler question: how are you thinking while you do it?

A surgeon can analyse tissue adversarially (looking for what is wrong) or engagingly (looking for what this pattern resembles across specialities). Same verb. Radically different cognition. The output changes even when the task stays the same.

This is not a minor gap. It is the reason your AI agents feel mechanical. They know what to do. They have no idea how to think while doing it.

⚛️ The Fusion

Three unrelated domains crash together here.

Cognitive science has long recognised that thinking has modes, not just levels. De Bono’s Six Hats are modes, not tasks. Kahneman’s System 1 and System 2 are postures, not skills. Jung’s four functions (Thinking, Feeling, Sensation, Intuition) are orientations, not actions. Donald Schon distinguished reflection-in-action from reflection-on-action: same mind, different gear. Michael Polanyi showed that tacit knowledge surfaces only through the act of making: you cannot articulate what your hands know until they do it.

AI agent architecture has ignored all of this. Modern agent systems define what an agent can do: search, write, analyse, debate. They are verb-based. The assumption is that the verb determines the cognition. It does not.

Living systems biology reveals why. An organism does not just perform metabolic functions; it adapts its posture to its environment. Fight-or-flight is not a new action. It is the same body in a different mode. The nervous system does not add new organs under threat; it reconfigures existing ones.

The collision: cognitive variation is orthogonal to cognitive action. You can perform any task in any mode. The mode changes the output even when the task stays the same.

Four Cognitive Gears Same task, different posture, different output

⚔️ ADVERSARIAL Posture: Test through opposition Question: "What could break this?" Core act: Friction Genuine when it produces unexpected findings. Not confirmation. BLIND SPOT: Never fuses (💥) 💥 ENGAGING Posture: Mesh across boundaries Question: "What engages across these?" Core act: Fusion Connections must be structurally isomorphic. Not merely superficial. BLIND SPOT: Never stress-tests (⚔️) 🪞 REFLECTIVE Posture: Mirror back, learn Question: "What did I learn?" Core act: Echo Exits when lesson is articulated AND committed to action. Otherwise: rumination. BLIND SPOT: Never ships (🔥) 🔥 GENERATIVE Posture: Create from constraint Question: "What can I make from these?" Core act: Flow Thinking occurs through making. Generation without judgment is spam. BLIND SPOT: Never pauses (🪞)

Any task × Any gear = Different output The variation is the posture, not the skill.

Four variations emerge from the collision:

VariationPostureCore ActWhen You Use It
Adversarial ⚔️Test through oppositionFrictionCode review, red-teaming, stress-testing a strategy
Engaging 💥Mesh across boundariesFusionCross-domain insight, pattern recognition, analogy
Reflective 🪞Mirror back, learnEchoPost-mortems, journaling, retrospectives
Generative 🔥Create from constraintFlowWriting, designing, prototyping under pressure

These are not tasks. They are cognitive gears. Any task can be run in any gear. An adversarial brainstorm produces different ideas than an engaging brainstorm. A reflective code review catches different bugs than an adversarial code review.

The New Pattern

This reframes how you design thinking architectures, human and artificial.

Old PatternNew Pattern
”The agent can search, write, and analyse""The agent can search adversarially, engagingly, reflectively, or generatively"
"Run a brainstorm""Run an engaging brainstorm (cross-domain fusion) or an adversarial brainstorm (red-team)"
"Perform a review""Reflective review (what did we learn?) or generative review (what could this become?)”
Define 12 specialised toolsDefine 4 variations across existing tools

For AI agents, this is a design lever. Instead of building 47 specialised tools, you build 4 cognitive variations across existing capabilities. The combinatorial space explodes without combinatorial code. Task times variation equals cognitive act: search:engaging, analyse:adversarial, write:generative, review:reflective.

For human teams, this is a facilitation lever. Instead of saying “let’s brainstorm,” you say “let’s brainstorm engagingly: bring one idea from a completely different field and let’s see where our domains engage.” The instruction changes the cognition.

The evidence is already in front of you: the SPAR protocol pioneered variation-awareness through its Style axis (adversarial, steelman, consensus). It was the first verb with a cognitive gear selector. The discovery is that every verb deserves one.

The Question

If cognitive variation is real, and any task can be run in any gear, then here is the question you carry forward:

Which gear do you default to, and what are you missing because of it?

The adversarial thinker never fuses. The engaging thinker never stress-tests. The reflective thinker never ships. The generative thinker never looks back.

Your default gear is your superpower and your blind spot. The question is whether you have ever named it.

cognitioncognitive_architecturevariationsworkflow_protocols