Capability System
What becomes possible when human judgment, machine leverage, and business demand are managed as one capability system?
A capability only matters when it changes an outcome. Human potential without a valued use can remain latent. An agent skill without doctrine becomes a prompt wrapper. A business capability without people or tools remains an aspiration.
The intersection is the unit to manage:
Human capability × Agent skill × Business capability → Proven outcome
Three Owners
Each domain keeps one job. Do not merge the three lists.
| Domain | Owns | Core question |
|---|---|---|
| Human capability | Character, know-how, judgment, trust, taste, and practice | What can a person understand, choose, and cultivate? |
| Agent skill | A repeatable method, tool use, gates, and execution receipt | What can AI do reliably within a declared boundary? |
| Business capability | The capacity to create, deliver, protect, and compound value | What must the business become able to do? |
The Capability System owns the edges between them: demand, delegation, overlap, evidence, maturity, gaps, and change over time.
Whole Capability
Cultivate every important capability with 100% stewardship. That does not mean every person masters every capability. It means no important capability is left ownerless, unpractised, unmeasured, or assumed.
Capability = Character + Know-how + Resources
- Character governs intent, responsibility, courage, trust, and restraint.
- Know-how turns information into situated judgment and repeatable action.
- Resources provide time, relationships, data, capital, tools, models, and machines.
AI can supply knowledge, synthesis, memory, and execution. It cannot silently inherit purpose, accountability, or the right to make a consequential promise.
Intersection Record
Use one record whenever a business capability is reviewed or an agent skill is
created. unknown is valid. An unsupported claim is not.
| Field | Required question |
|---|---|
| Business outcome | Which valuable change or promise must the business produce? |
| Human capability | Which judgment, character, relationship, or physical presence is required? |
| Agent contribution | Which repeatable work can an agent perform, check, or coordinate? |
| Delegation state | Is the human doing, assisted, orchestrating, automating, or governing? |
| Unit of Work | What is the smallest valuable action at this intersection? |
| Proof signal | What observable output and outcome show the combination worked? |
| Evidence state | Is the relationship unknown, hypothesised, observed, tested, or proven? |
| Maturity | Is it an idea, defined, connected, tested, or proven capability? |
| Owner | Who cultivates the human edge and who owns the business result? |
| Freshness | When was the evidence last checked, and what could invalidate it? |
| Next question | Which uncertainty should the next experiment reduce? |
The delegation state uses one ladder:
Do → Assist → Orchestrate → Automate → Govern
Move right only when proof improves and risk remains controlled. Automation is not maturity when a human must repair every important exception.
Seed Intersections
These rows demonstrate the record. They are hypotheses until a named business context supplies a baseline and receipt.
| Business need | Human edge to cultivate | Agent contribution | Current evidence | Next proof |
|---|---|---|---|---|
| Understand demand | Listening, empathy, questioning | Gather signals, cluster language, test contradictions | observed | Compare an agent-assisted interview synthesis with the buyer's correction |
| Choose direction | Critical thinking, planning, courage | Build options, surface assumptions, preserve the decision trace | observed | Measure decision reversals caused by evidence found before commitment |
| Earn trust | Writing, storytelling, taste | Research, draft, check claims, adapt formats | observed | Test whether sourced agent-assisted work improves qualified action without weakening voice |
| Deliver reliably | Focus, orchestration, leadership | Route work, monitor gates, expose variance | hypothesised | Run one bounded workflow with named human escalation and outcome receipt |
| Improve the system | Pattern recognition, truth recognition, optimism | Compare setpoint with output, retain lessons, propose the next question | observed | Show that the next run detects or prevents the previous variance earlier |
Five-Step Loop
- Demand — name the business outcome, beneficiary, urgency, baseline, and owner.
- Decompose — identify the human capability, repeatable work, resources, and consequential judgment.
- Delegate — place each action on the delegation ladder and set escalation boundaries.
- Demonstrate — run the smallest safe experiment and capture output, outcome, intervention, and exception evidence.
- Evolve — update the relationship, maturity, freshness, next question, and rightful Pattern, Process, or Platform owner.
Review high-consequence and fast-changing intersections quarterly. Review a stable proven intersection annually. Review immediately when a model, tool, law, business promise, failure mode, or human role materially changes.
Portfolio Gauge
Measure the system as coverage before scoring quality.
- Demand coverage: business capabilities with a named outcome and owner.
- Human coverage: outcomes with a named human edge and cultivation practice.
- Agent coverage: repeatable work with a skill, tool, boundary, and receipt.
- Proof coverage: intersections with a baseline, experiment, outcome, and evidence state.
- Freshness coverage: intersections reviewed within their declared interval.
Missing coverage remains unknown. It does not become a weak human, weak
agent, or weak business score.
Run It
Copy this prompt for one capability review. Keep the unit small enough to prove.
Map one capability intersection.
Business outcome and beneficiary:
Current workflow and baseline:
Human capabilities used today:
AI agents, skills, tools, and data available:
Consequential decisions and risks:
Evidence cutoff:
Return:
1. the smallest valuable Unit of Work;
2. the human edge that must be cultivated;
3. the agent contribution and delegation state;
4. overlaps, gaps, and unsafe substitutions;
5. one bounded experiment with owner, setpoint, proof, stop condition, and next question.
Separate observed evidence, inference, and unknowns. Do not equate automation
with capability. Change the map only when the experiment changes an outcome.
Evolution Rule
The tactical question is: Which missing or weak intersection is constraining the outcome now?
The strategic question is: Are we cultivating human agency and business capacity, or merely increasing machine activity?
Promote a relationship into a reusable standard only after more than one context produces the same mechanism and proof. Retire it when the business need disappears, the agent skill no longer changes work, or the claimed human edge does not survive evidence.
Context
- Folder Ownership — the folder → capability → Unit of Work → skill → tool → evidence chain
- Agency Capabilities — the human capability portfolio and cultivation doctrine
- Agent Skills — repeatable AI-enabled methods and maturity
- Business Capabilities — the functions that create, deliver, and protect value
- Work Charts — activity-level human and AI allocation
- Evolution — how measured variation changes the next system
Questions
Which business outcome is currently limited by a missing human capability, an immature agent skill, or a poorly designed intersection between them?
- Which capability are we automating before we understand it?
- Which human capability becomes more valuable when AI handles the volume work?
- Which agent skill produces activity without a business outcome receipt?
- Where is one capability duplicated across teams because no shared owner exists?
Next question: which intersection should become the first quarterly capability experiment?