Tools and Skills Matrix
Which of your tasks is a human doing only because nobody wrote it down precisely enough to hand over?
A capability list tells you what a business can do. A tool list tells you what it owns. Neither tells you who or what should perform a given piece of work — and that is the decision that actually moves cost, speed, and quality. This matrix answers it by laying one axis against another: the function's tasks down the side, the things that could perform them across the top.
The rows are not invented here. They come from the process pillar of each function, where Process Inventory has already decomposed the work into verb-led task statements with stable IDs. That is the precondition. Work that was never written as a task cannot appear in this matrix, which is why it is never consciously allocated, delegated, or automated — it just quietly stays with whoever has always done it.
The Columns
Six columns, and every one of them already has a canonical home. This page points at those homes; it does not restate them. A matrix that re-lists a taxonomy becomes a second source of truth, and the moment the two disagree the matrix is the one that is wrong.
- Task — the row itself.
function.verb-object.Tnn, read from the job page in that function's process pillar. - Player — who holds it. Defined in the function's players pillar; society-level actors sit in Ecosystem, internal roles in Roles.
- Human or agent — the split, and where human authority is retained. Framed by the function's work chart.
- Skill — what capability the task demands. For agent work, classify with the TACO framework — Tasker, Automater, Collaborator, Orchestrator — at AI Integration Strategy.
- Tool — what it runs on. AI Toolkit for agent tooling, and the function's platform pillar for everything else.
- Evidence of done — the observable signal the task completed. Sourced from the job page, not chosen here.
Where the task's input or output is not text, add the modality from Modalities — it usually decides whether a tool can do the task at all. Where the question is which market a capability serves rather than who performs it, that is industry and function fit, and it belongs in Function Catalog.
Reading One Row
| Task | Player | Human or agent | Skill | Tool | Evidence of done |
|---|---|---|---|---|---|
sales.qualify-leads.T01 Scores each inbound lead against the ideal customer profile | Qualification agent | agent scores | Tasker | CRM | Scored lead with reason |
sales.qualify-leads.T03 Confirms or overrides the score where the case sits on a boundary | Sales lead | human decides | Fit judgment | CRM | Recorded decision with rationale |
sales.qualify-leads.T04 Records the disqualification reason for every rejected lead | Qualification agent | agent records | Tasker | CRM | Reason field populated |
Three rows from one job, and they allocate differently. T01 and T04 are mechanical and observable, so an agent can hold them. T03 is a judgment on a boundary case, so a human holds it — not because agents cannot score, but because the cost of a confident wrong answer lands on a relationship. The split falls out of the task statements. It was not negotiated.
Finding An Automation Candidate
A task is a genuine candidate when three things are true at once:
- The statement is observable. A verb-led statement naming a concrete object can be checked. A topic label cannot, so it cannot be handed over.
- The evidence of done is machine-checkable. If completion is a field populated or an artifact produced, an agent can close the loop. If completion is "the client felt heard", it cannot.
- No irreversible judgment sits inside it. Where a wrong answer costs money, trust, or a commitment that cannot be withdrawn, a human keeps the decision even when the agent does the work leading up to it.
All three, or it is not a candidate yet. Two out of three usually means the task needs rewriting before it needs a tool.
Honest Gaps
Two rules keep this matrix from flattering the business it describes.
A named tool or skill with no canonical page is a gap, not a resolved cell. Write it as a gap. The moment a matrix fills a cell with a tool that has no documented home, it starts asserting capability the business has not shown. The gap is the useful output: it names what to document or acquire next.
An aggregate is coverage, not performance. A figure like "47% AI share" says how much of the written work is allocated to agents on a given date. It says nothing about whether the function is performing better. Those are separate claims with separate evidence, and collapsing them is the most common way an AI-transformation number becomes a lie. Date every aggregate, and keep outcome claims in the function's performance pillar where they can be gauged.
Where The Rows Come From
| Function | Inventory | Task rows available |
|---|---|---|
| Sales | Sales Process | 5 jobs decomposed, 16 jobs declared as gaps |
| Marketing | Marketing Process | 8 jobs declared, none decomposed to tasks |
| Every other function | Business Capabilities | Inventories not yet built |
| Industries | Industries | Process pillars are single pages, not inventories |
That table is the honest state of this matrix: one function has task rows, one has job rows, and the rest have neither. The matrix is only as complete as the inventories feeding it, and building those is the work described in Process Inventory.
Context
- depends-on Process Inventory — produces the task rows this matrix reads.
- depends-on Information Architecture Standard — why the process pillar decomposes to the task level.
- pairs-with AI Toolkit — the canonical home of the tool column.
- pairs-with Function Catalog — industry and function fit, a different question from who performs the task.
- pairs-with Modalities — the input and output constraint that often decides tool feasibility.
- applies-to Sales Process — the one inventory with task rows today.
- contrasts-with Select Tech — this matrix says what work needs a tool; that method decides which tool to adopt.
- risk-governed-by Delegation — what authority may be handed to an agent, and what may not.
Questions
Next question: which task in your function fails the automation test only because its statement is a label rather than a verb?
- Which cell in your matrix names a tool that has no documented home?
- Which human-held task is held by habit rather than by judgment?
- What would your aggregate figure be if every undocumented job were counted as unallocated?