An action is an intervention: something an agent does that changes, preserves or attempts to influence a state.
Teaching, approving, repairing, measuring, writing, purchasing, moving, deleting, regulating, voting and warning are all actions. What matters is not only the verb, but who acts, why, by what method, under what authority, on what target, with what effect and how reversible the change is.
Quick answer: how should actions be categorised?
- Intent: what is the actor trying to achieve?
- Method: how is the action performed?
- Actor: who or what acts?
- Authority: is the actor permitted to act?
- Target: what does the action affect?
- Scope: local, system-wide, temporary, permanent?
- Timing: immediate, scheduled, recurring, conditional?
- Reversibility: can the action be undone?
- Effect: what state changes directly?
- Consequence: what happens afterward?
This article applies How to Categorise Anything to actions as intentional or automated changes in the world.
1. Separate action from event
An action has an agent performing it. An event may occur without intentional agency.
2. Separate action from process
An action is usually one step; a process is an organised sequence of actions and state changes.
3. Intent is one axis
Prevent, create, inspect, move, repair, communicate, decide and enforce are different action purposes.
4. Intended and actual effect can differ
An action can fail, overshoot or create side effects. Preserve intent separately from observed consequence.
5. Actor identity matters
Person, organisation, machine, software agent and autonomous controller are different actor types.
6. Role matters too
The same person can approve in one role and observe in another.
7. Authority should be explicit
Permitted, delegated, unauthorised and emergency actions have different governance meaning.
8. Target defines what changes
An action may target an object, person, system, process, document, resource or decision state.
9. Method is separate from purpose
Two actions can pursue the same goal through manual, automated, physical, digital or procedural methods.
10. Direct and indirect actions differ
Some actions alter the target directly; others change incentives, information or constraints around it.
11. Preventive actions precede failure
Inspection, maintenance and vaccination are examples where action aims to reduce future risk.
12. Corrective actions respond after deviation
Repair, rework and correction attempt to restore an expected state.
13. Detective actions produce information
Testing, monitoring and auditing may not change the system directly but change what is known.
14. Escalation actions change authority level
Escalation routes a case to a role with broader powers or expertise.
15. Scope should be measured
A local patch and a system-wide migration may use similar verbs while having very different blast radius.
16. Reversibility matters
Undoable actions tolerate experimentation better than irreversible actions.
17. Reversibility can be partial
A record can be restored while lost trust or elapsed time cannot.
18. Timing can be immediate
Emergency shutdowns and alerts act as soon as a trigger is detected.
19. Timing can be scheduled
Maintenance, publication and reporting actions may occur at planned times.
20. Timing can be conditional
An action may occur only when a threshold, event or approval condition is met.
21. Repeated actions form routines
Recurring actions can become process steps or automated workflows.
22. Automated actions need provenance
Record which rule, model or system initiated consequential automated actions.
23. Human approval can be a separate action
Recommendation, approval and execution should not be collapsed into one step.
24. Action legality is contextual
The same physical act can be lawful, restricted or prohibited depending on jurisdiction and authority.
25. Ethical status is not identical to legal status
Keep evaluative and legal classifications separate where both matter.
26. Effect should be observed
Record what changed immediately after the action rather than assuming intent became reality.
27. Consequences can be delayed
Some effects appear later or propagate through dependencies.
28. Side effects deserve representation
An action can succeed on its primary target while creating new problems elsewhere.
29. Action chains need relationships
Triggered-by, enables, blocks, follows and compensates-for are useful relations among actions.
30. Compensating actions restore balance
When an original action cannot be reversed directly, a later action may offset its effect.
31. Action failure has types
Not executed, partially executed, wrongly targeted, mistimed and ineffective require different responses.
32. Evidence of execution matters
Logs, signatures, sensor traces and receipts can show that an action actually occurred.
33. AI can classify actions from language
Models can extract actor, verb, target and intent, but inferred intent should remain distinguishable from explicit evidence.
34. Action taxonomies drift
New technologies create new forms of automation and interaction that older action schemes may not cover.
35. Action state is separate from action type
Planned, approved, executing, completed, failed and cancelled describe lifecycle.
36. Actions should connect to decisions
Many actions are consequences of decisions, while others generate evidence for future decisions.
37. A practical action record
- action ID;
- action type;
- intent;
- actor and role;
- authority;
- target;
- method;
- scope;
- timing;
- reversibility;
- expected effect;
- observed effect;
- consequences;
- evidence;
- status;
- version.
38. Classify action by what changes
The same verb can describe different actions when target, authority or effect differs.
39. Action classification should support accountability
A useful scheme can answer who acted, under what authority, on what target and with what result.
40. The deeper idea
Actions are the bridge between decision and changed reality.
To categorise an action well is to preserve intention, agency, authority, target and consequence in the same frame.
Final answer
Categorise actions by intent, method, actor, authority, target, scope, timing, reversibility, effect and consequence. Separate action from event and process, intended from observed effect, and approval from execution. Preserve evidence and provenance for consequential actions.