Routines and Execution Model
ThinkWork documents routines as multi-step automations that can chain agent invocations, branching, and external actions.
Execution model
Section titled “Execution model”All automations are backed by AWS Step Functions.
A typical run looks like this:
- EventBridge or an API trigger starts the state machine
- ThinkWork creates an
AUTO-thread for the run - AgentCore executes the configured step or prompt
- The run finishes as
succeededorfailed
For routines, each step becomes part of the Step Functions execution path.
Runs and debugging
Section titled “Runs and debugging”Every automation run is recorded with status, timing, and the associated thread.
That gives you two useful views of the same work:
- the thread view, which shows the agent conversation and outputs
- the execution view, which shows state-machine progress and failures
Manual triggers and retries
Section titled “Manual triggers and retries”Automations can also be triggered manually, and retry behavior can be configured for failed runs.
That makes it practical to:
- test before enabling a schedule
- rerun a failed job
- override the prompt for a one-off run