Build familiar, rule-based workflows with Pathways. Use Thinkways to structure reasoning and work stages inside an AI agent—not to create another agent. Agent Studio sets the agent's autonomy and permissions.
How the pieces fit together
Pathway · defined workflow
Receive request → Load permitted data
Thinkway · method inside the agent
A repeatable approach to a changing case.
01
Clarify the request and collect context.
02
Assess evidence against the method.
03
Prepare the next permitted action.
Each node describes a work stage: instructions, tools, skills and conditions.
Policy and approval check
Execute the permitted action → Record the result
Illustrative workflow, not a live execution.
Pathways are rule-based workflows with predefined steps, conditions and transitions—the familiar workflow-automation model. Use them for known transformations, routing and approved actions.
Thinkways are reusable reasoning workflows inside AI agents, not agents themselves. Their nodes describe work stages with instructions, tools, skills and entry conditions, giving the agent a method to apply to changing context.
Connect steps, conditions, tools and approval gates in a familiar workflow. Define the route explicitly; use an agent only where the work needs interpretation.
Give an agent a reusable approach to research, qualification, diagnosis or review. Nodes organise instructions, resources and conditions around each stage; the agent still does the reasoning.
If the business already knows the correct sequence, turning every step into an open-ended agent decision adds variance without adding intelligence. Deterministic execution remains the better primitive for fixed transformations, validations, routing rules and known approval chains.
Createting's architecture keeps Pathways available precisely so teams can reserve reasoning for the parts of a workflow that genuinely require it.
Research, diagnosis, exception handling and ambiguous customer requests can require interpretation that is expensive or impossible to enumerate up front. Forcing every possibility into a static graph produces brittle automation and endless edge-case branches.
A Thinkway makes the method reusable, not every answer identical. Nodes describe stages such as gather, assess and verify; the agent applies that method to the case at hand. Model outputs can still vary.
A hybrid design lets deterministic infrastructure prepare the problem, reasoning handle ambiguity, and deterministic execution take over again once the decision is clear.
Example: a Pathway loads approved account data. A support agent uses a Thinkway to diagnose, check evidence and propose a resolution. A commercial concession requires approval; the Pathway then performs the permitted update.
Actor: support agent · Method: Thinkway · Process: Pathway
Example: a Pathway loads company records. A research agent follows a Thinkway to gather evidence, assess fit and verify a qualification brief. The Pathway routes the result to the appropriate sales process.
Actor: research agent · Method: Thinkway · Process: Pathway
Example: a Pathway collects system state. An operations agent uses a Thinkway to investigate, compare recovery options and verify its proposal. High-impact actions require approval before the Pathway executes the permitted route.
Actor: operations agent · Method: Thinkway · Process: Pathway
Agent Studio configures agent autonomy. A Thinkway structures the method; its mode specifies required guidance, suggested guidance or reference use. Action permissions and approvals are separate runtime controls—not authority granted by a Thinkway.
Where does the Thinkway fit? When the agent should apply a reusable professional method, attach a Thinkway to that agent. The Thinkway structures how the agent works; it does not replace the agent.
A Pathway is a rule-based workflow with explicit steps and transitions. A Thinkway is a reusable reasoning workflow inside an AI agent. Its nodes organise work stages, instructions, tools, skills and conditions. The agent performs the work; the Thinkway structures its approach.
A Pathway follows a defined workflow with explicit steps and transitions. An AI agent interprets context and performs work within its configured autonomy and permissions. When that agent should apply a reusable method, a Thinkway structures its reasoning and work stages inside the agent.
A hybrid process can use a Pathway for known steps and delegate interpretive work to an agent using a Thinkway. The agent applies its method within configured permissions, then returns a result so structured execution can continue. A Thinkway is not a separate worker.
Because fixed work benefits from repeatability. Reasoning adds value where interpretation is required, but it also introduces variance that is unnecessary for known transformations and routes.
A Thinkway is not an authorisation mechanism. The agent's configured autonomy, tool policies and approval requirements govern actions. Changing Thinkway mode does not itself authorise an external action.
Use a Pathway for predictable stages and an agent with a Thinkway for stages that require interpretation. Thinkways can also describe reusable professional methods for research, qualification or review—not just exception handling.
Nodes describe work stages or states inside the agent's reasoning workflow. A stage can group instructions, tools, skills and entry conditions; edges connect the stages. This makes the working method explicit. It is not a view into a model's private chain of thought, and a node is not another agent.
No. Agent Studio configures the agent's autonomy and permissions. A Thinkway structures its working method and can be used with human supervision or greater permitted autonomy. FORCED, SUGGESTED and REFERENCE describe how the method is used, not whether the agent may bypass action policy or approvals.
The configured method: its stages, instructions, resources and review criteria. Teams can reuse a professional approach across cases instead of describing it from scratch. Model outputs can still vary; the required results and controls should be tested for the intended workflow.
Map one real workflow before choosing the architecture.