Capabilities, Skills and Tools

Knowing something and being able to do something are not the same. Neither is teaching a human and teaching an AI.

The Capability Gap

Here’s a frustrating reality: your senior expert can do remarkable things, but she may not be able to explain how. And even if she could explain it perfectly, that explanation might not help an AI system do the same thing.

This is the capability gap — the space between human expertise and machine execution.

Closing it requires understanding three distinct layers: capabilities, skills, and tools. Each works differently for humans and AI systems. Most knowledge engineering efforts fail because they conflate these layers.

Three Layers, Two Worlds

Capabilities

What outcomes can be achieved. The “what.”

For humans: Capabilities are often implicit. A skilled negotiator “can close complex deals.” We don’t decompose this further.

For AI: Capabilities must be explicitly defined and bounded. The system needs to know exactly what it’s supposed to achieve — and what’s out of scope.

Skills

What competencies enable the capability. The “how.”

For humans: Skills develop through practice, feedback, and experience. They’re often unconscious — experts don’t think about the component skills anymore.

For AI: Skills must be decomposed into discrete, teachable components. Each skill requires specific knowledge, data access, or training.

Tools

What instruments are used to execute. The “with what.”

For humans: Tools are extensions of skills — a spreadsheet, a CRM, a phone. Experts choose tools fluidly based on context.

For AI: Tools are precisely defined interfaces with clear inputs, outputs, and constraints. An agent can only use what it’s been given access to.

graph TB
TITLE["CAPABILITY DECOMPOSITION"]

TITLE --> CAP["<b>CAPABILITY</b><br/>Handle customer credit decisions"]

CAP --> SKILLS["<b>SKILLS</b>"]

SKILLS --> S1["Risk<br/>Assessment"]
SKILLS --> S2["Policy<br/>Application"]
SKILLS --> S3["Exception<br/>Handling"]

S1 --> TOOLS["<b>TOOLS</b>"]
S2 --> TOOLS
S3 --> TOOLS

TOOLS --> T1["CRM<br/>Lookup"]
TOOLS --> T2["Credit<br/>Scoring"]
TOOLS --> T3["Policy<br/>Engine"]
TOOLS --> T4["Escalation"]
TOOLS --> T5["Communication"]

T1 --> KR["<b>KNOWLEDGE REQUIREMENTS</b><br/>• Customer history (DATA)<br/>• Credit policies (RULES)<br/>• Exception criteria (JUDGMENT)<br/>• Escalation triggers (BOUNDARIES)"]
T2 --> KR
T3 --> KR
T4 --> KR

style TITLE fill:#fef2f2,stroke:#dc2626,stroke-width:4px,color:#991b1b,font-size:18px
style CAP fill:#fef2f2,stroke:#dc2626,stroke-width:3px,color:#991b1b,min-width:300px
style SKILLS fill:#fff7ed,stroke:#f97316,stroke-width:3px,color:#c2410c,min-width:200px
style S1 fill:#fff7ed,stroke:#f97316,min-width:120px
style S2 fill:#fff7ed,stroke:#f97316,min-width:120px
style S3 fill:#fff7ed,stroke:#f97316,min-width:120px
style TOOLS fill:#fef2f2,stroke:#dc2626,stroke-width:3px,color:#991b1b,min-width:200px
style T1 fill:#fef2f2,stroke:#dc2626,min-width:100px
style T2 fill:#fef2f2,stroke:#dc2626,min-width:100px
style T3 fill:#fef2f2,stroke:#dc2626,min-width:100px
style T4 fill:#fef2f2,stroke:#dc2626,min-width:100px
style T5 fill:#fef2f2,stroke:#dc2626,min-width:100px
style KR fill:#fff7ed,stroke:#f97316,stroke-width:3px,color:#c2410c,min-width:350px

Why This Decomposition Matters

When you want an AI system to take over (or augment) a human capability, you need to:

1. Identify which skills the capability actually requires — often more than you’d expect
2. Determine which skills can be encoded vs. which need human oversight
3. Ensure the right tools are available and accessible
4. Map the knowledge required at each layer

Most AI projects fail at step one. They try to replicate a capability without understanding the underlying skills. This is like trying to teach someone to “be a good manager” without teaching delegation, feedback, prioritization, or any other component skill.

The Corporate Executive’s Action Framework

Translating this into organizational action requires a structured approach:

Step 1: Capability Mapping

What to do: Inventory the capabilities your organization depends on. Not job titles or departments — actual outcomes.

Example capabilities:

• Accurately price complex custom orders
• Identify compliance risks before they escalate
• Onboard new clients without losing context
• Resolve customer complaints that deviate from standard policy

Key question: If this capability disappeared tomorrow, what would break?

Step 2: Skill Decomposition

What to do: For each critical capability, identify the component skills. Be specific.

Example for “Price complex custom orders”:

• Interpret technical specifications
• Assess manufacturing complexity
• Apply pricing matrices with judgment
• Factor in customer relationship context
• Identify margin risks
• Communicate rationale clearly

Key question: Which of these skills could a well-designed system do? Which require human judgment?

Step 3: Tool Inventory

What to do: Map what tools exist, what data they access, and whether AI systems could use them.

Key question: Where are the gaps between human tool access and AI tool access?

Step 4: Knowledge Gap Analysis

What to do: Identify what knowledge is missing at each layer.

Key question: What would need to be true for an AI system to execute this skill?

The Build-Buy-Partner Decision

Once you’ve mapped capabilities, skills, and tools, you face a strategic choice:

Build internally: When the capability is core to your competitive advantage and the knowledge is highly proprietary.

Buy off-the-shelf: When the capability is generic (scheduling, basic support) and your context doesn’t fundamentally change how it works.

Partner for extraction: When the capability is critical, the knowledge exists but isn’t captured, and you need specialized methods to surface it.

Most organizations need all three. The mistake is defaulting to “buy” for everything and wondering why the AI doesn’t understand their business.

What Good Looks Like

Organizations that get this right share patterns:

They think in capabilities, not features. Instead of “we need a chatbot,” they ask “what capability should this enable, and what skills does that require?”

They decompose before they automate. Before deploying AI, they map the skill and knowledge requirements. This often reveals that automation is premature — the knowledge isn’t ready yet.

They build tool bridges deliberately. They create APIs, structured outputs, and integration layers that let AI systems access what they need — without compromising security or governance.

They preserve human judgment where it matters. Not everything should be automated. The goal is a system where AI handles the repeatable and humans handle the exceptional.

The Uncomfortable Question

Most organizations think their challenge is “we need better AI tools.”

The real challenge is usually: Do we actually understand our own capabilities well enough to teach them to anyone — human or machine?

If you can’t decompose a capability into skills, you can’t transfer it to a new employee effectively either. AI just makes this gap more visible.

Want to go deeper?

This decomposition is exactly what happens in a Design Workshop. We map your critical capabilities, identify the skills and knowledge they require, and design an extraction approach that makes your expertise teachable — to humans and AI alike.

Book a Design Workshop →