Is agentive logic one fixed logic?

No. It is an area containing related formal approaches to agency and action.

Is STIT the name of an AI assistant?

No. It abbreviates “seeing to it that” in a family of logical treatments of agency.

Does a formal model decide who is legally responsible?

No. It can clarify assumptions and relationships, but legal responsibility requires the relevant rules and facts.

AI Glossary / AI and machine learning / Advanced

Agentive logic

Definition

Agentive logic is a family of formal approaches for reasoning about what agents do, can bring about or are responsible for bringing about. It distinguishes an outcome occurring from an agent seeing to it that the outcome occurs.

What it means

Suppose a room becomes quiet. That fact alone does not tell us whether someone deliberately stopped a machine, whether its timer expired or whether the power failed. Statements about action contain information that an ordinary description of the final state leaves out.

Agentive logic provides a precise language for examining that difference. An agent can be a person, an organization or a software actor within the model. The term describes a subject of logical study, not a particular chatbot product or a claim that machines have human intentions.

For business readers, its main value is conceptual clarity. A workflow may need to distinguish being able to perform an action, being permitted to perform it, attempting it and actually ensuring the desired result. Those are different claims.

How it works

A formal account specifies possible situations, the choices available to an agent and the outcomes associated with those choices. One influential approach uses “seeing to it that,” commonly abbreviated STIT, to express agency.

Consider a simplified choice between locking and leaving a door unlocked. If every outcome compatible with the locking choice includes a locked door, the model can represent the agent as ensuring that condition. If a broken mechanism may leave the door open, selecting the command is insufficient to guarantee the result.

Different systems represent agency, time and alternatives differently. Extensions can examine groups, obligations or responsibility. They require explicit definitions: a logical symbol does not supply a real-world account of responsibility by itself. When such ideas inform software, engineers still need to connect abstract choices to permissions, commands, observations and failure handling.

Three practical examples

A workflow analyst distinguishes an assistant requesting a refund from an authorized service actually ensuring that the refund is issued.
A multi-agent researcher represents whether two cooperating agents can jointly achieve a goal that neither can ensure alone.
A policy designer separates what an automated scheduler can arrange from what organisational rules permit it to arrange.

Think of it this way

Think of a stage manager who can switch on a spotlight. The stage becoming bright is an outcome; the manager choosing a working lighting control is an action that can ensure it. Sunlight through a window might produce the same outcome without that action.

The analogy is deliberately small. Real responsibility may involve intentions, knowledge and social rules. A simple action model does not settle those wider questions.

Worked example

Separating a command from control of the result

In an invented file-sharing workflow, an agent has two choices: request restricted access or request public access. The storage service may either apply the request or return an error.

There are therefore four simple choice-and-response combinations. Of the two combinations involving a restriction request, only one ends with confirmed restricted access. Under this model, the agent's initial choice does not ensure that the file becomes restricted.

The designer changes the task definition to “request restriction, check the returned state, and escalate if restriction is not confirmed.” That sequence can satisfy a reporting obligation even when the agent cannot guarantee the storage outcome. The example is a reasoning exercise, not a mathematical proof of a deployed service. Its purpose is to expose the difference between an attempted action and an ensured condition.

Illustrative case study

A fictional training provider designs several software agents to organize workshops. One agent selects dates, another reserves rooms and another sends invitations. The initial specification says that the date-selection agent “ensures a room is available.”

A review of the available actions shows that this agent can only ask the booking agent for a reservation. The team rewrites the responsibilities: date selection produces proposals, booking confirms availability, and invitations require that confirmation. It also defines what happens when a booking expires.

The improvement comes from separating each actor's choices and guarantees. No claim is made that the business implemented a full agentive-logic theorem prover.

Common misconceptions and mistakes

Any successful outcome proves that an agent caused it. The same outcome may occur independently of the agent's choice.
Ability and permission are interchangeable. An action can be technically possible while disallowed by a rule.
Agentive logic proves that an AI is conscious. Formal models of action do not establish subjective experience.

Business use

Most managers will not write formulas in an agentive logic. They can still use its distinctions when assigning responsibility across automated workflows. Ask which actor controls each decision and which actor merely requests or observes an outcome.

This is particularly useful when several services share a task. Clear action boundaries support better exception handling and reduce vague statements such as “the AI will take care of it.” Logical modeling is an optional technical method for making selected relationships precise.

Check your understanding

A booking assistant requests a room, but the booking service can reject the request. Has the assistant ensured that the room is booked?

Answer: Not merely by requesting it. The model must account for the service response and any further actions that establish confirmation.

Related terms

Agent
Modal logic
STIT logic
logic of action
multi-agent system
planning
formal verification

Learn next

Read logic of action and state machine. Choose a shared workflow and list what each participant can request, what it can confirm and what it can independently ensure.

Sources

Stanford Encyclopedia of Philosophy, The Logic of Action: https://plato.stanford.edu/entries/logic-action/. Scholarly reference for approaches to action and STIT. Examples are original simplified teaching models; they do not constitute legal or philosophical conclusions about a real system.

Previous: Agentic workflow Next: Agglomerative clustering