The Four Pillars (Augmentation, Insights, Orchestration, Generation)

Definition

The Four Pillars are the “what AI does” tier of the AI Capability Framework created by Greg Kihlström in his book Stop Saying “AI”. They name the four distinct things AI does, so a conversation about “AI” can become a conversation about a specific capability with its own owner, cost, and risk.

• Augmentation. AI as a capability amplifier for a human author. It closes gaps in speed, skill, or knowledge while the person stays in the authoring seat. This is the broadest pillar by volume, and it deliberately absorbs most of what people loosely call “AI creativity.” The output is a person’s work, done better or faster.
• Insights. AI as a sense-making engine — pattern detection, prediction, explanation, surfacing the non-obvious from data. The output is understanding, which then feeds a decision.
• Orchestration. AI as coordinator and executor across systems, tools, and other agents. It spans the full supervision spectrum, from human-directed workflows at one end to unsupervised autonomous execution at the other. Autonomy isn’t a separate category — it’s simply the unsupervised end of this same dial. The output is action and coordination.
• Generation. AI as artifact synthesizer, where the artifact itself is the unit of value and a human is not necessarily the author. The output is the thing made.

A key property: the pillars are lenses that can stack, not bins that must be mutually exclusive. Most real experiences are more than one pillar at once. The everyday “AI made me a thing” is usually Generation deployed in service of Augmentation, where the two pillars combine. The pillars also say nothing about how a capability is built — that’s the separate mechanism axis, which crosses all four.

How It Relates to Marketing

Nearly every AI feature in a marketing stack resolves into one or more of these four pillars, and naming which one changes the conversation from vague to fundable.

• Augmentation covers the copilots: drafting, rewriting in a brand voice, summarizing a call, coaching a rep. A marketer stays the author.
• Insights covers lead scoring, propensity models, attribution, segmentation, and anomaly detection. The output is a sharper decision, not a finished asset.
• Orchestration covers journey execution, next best action, triggered sends, and increasingly agents that act across tools. The output is coordinated activity.
• Generation covers producing creative at a scale or dimensionality no human authoring loop would reach — thousands of variants, synthetic test data, procedurally generated assets.

Sorting the stack this way exposes two things quickly: capabilities that are quietly doing high-stakes work under a low-stakes label, and capabilities that are duplicated across three vendors because nobody named the pillar they share.

The Four Pillars at a Glance

Pillar	What it produces	Who authors	Marketing example
Augmentation	A person’s work, amplified	The human	Draft assist, call summaries, brand-voice rewriting
Insights	Understanding that feeds a decision	The human decides	Lead scoring, attribution, anomaly detection
Orchestration	Action and coordination	Shared, across a supervision dial	Journey execution, triggered sends, agents
Generation	The artifact itself	Not necessarily a human	Variant generation at scale, synthetic data

How to Sort a Capability into a Pillar

1. Ask what the output actually is. Understanding points to Insights. A finished artifact points to Generation or Augmentation. Coordinated action points to Orchestration.
2. For anything that makes something, run the Author Test. If a skilled human could still produce it unaided, just slower, it’s Augmentation. If the output exists only because AI synthesizes at a scale no human loop could reach, it’s Generation.
3. Don’t force a single pillar. If a capability is genuinely two — a generation engine serving a human author — name both. The pillars stack.
4. Name the pillar before the mechanism. Decide what the capability does before arguing about whether it needs a generative model or a rule.

Comparison to Similar Frameworks

Framework	Organizes by	Difference
The Four Pillars	What AI does (capability)	Capability-first; mechanism-agnostic
Descriptive / Predictive / Prescriptive Analytics	Analytical maturity	Sits inside the Insights pillar
Automation vs. autonomy debates	Degree of human control	Folded into Orchestration’s supervision dial
Generative AI as a category	The underlying technique	A mechanism, not a pillar — see the two collisions

The pillars differ from most AI taxonomies by refusing to organize around the technology. Analytics maturity, automation levels, and “generative vs. traditional” all describe how something is built. The pillars describe what it does, which is the thing a leader budgets and governs.

Best Practices

• Name the pillar first, always. Capability before mechanism, before vendor.
• Let pillars stack. Treating them as mutually exclusive bins forces a false choice and hides the common “generation serving augmentation” case.
• Watch the two collisions. A generative mechanism is not the Generation pillar; an agentic mechanism is not the Orchestration pillar. The shared vocabulary causes real over-buying.
• Keep Insights honest about its output. Insights produces understanding, not decisions. The human still decides, which is where trust and calibration matter.

Future Trends

• Orchestration grows toward the unsupervised end. As agents mature, more capability slides down the supervision dial, which raises the stakes on the Trust and Identity layers rather than creating a new pillar.
• Generation and Augmentation blur in the product, stay distinct in management. Consumer features will feel like one thing while still needing to be governed as two.
• Insights re-centers on decisions, not dashboards. As understanding gets cheap, the scarce step becomes acting on it — which hands back up to the goals apex.

FAQs

1. Where do the Four Pillars come from? They’re the “what AI does” tier of the AI Capability Framework in Stop Saying “AI” by Greg Kihlström.

2. Are the pillars mutually exclusive? No. They’re lenses that stack. Most real experiences are more than one pillar at once, most commonly Generation in service of Augmentation.

3. Isn’t “Generation” just generative AI? No, and this is the most common error. Generative AI is a mechanism (how). The Generation pillar is a capability (authorless synthesis at superhuman scale), which can even be delivered without a generative model.

4. Where does autonomy fit? Inside Orchestration. Autonomy is the unsupervised end of Orchestration’s supervision dial, not a separate pillar.

5. What about AI that helps set strategy? That’s Insights and Augmentation applied to the strategy domain — a use case inside existing pillars, not a fifth pillar. Goals stay at the apex of the framework.

Related Terms

1. The AI Capability Framework (Master Stack)
2. Augmentation vs. Generation
3. The Author Test
4. The Mechanism Axis
5. The Supervision Dial
6. Generative AI
7. Agentic AI
8. Predictive Analytics
9. Next Best Action (NBA)
10. Customer Journey Orchestration (CJO)

Sources

• Kihlström, Greg. Stop Saying “AI”. https://amzn.to/4wilWcA
• Greg Kihlström — official site. https://www.gregkihlstrom.com