What changed?
A new source reading is waiting to be understood.
Guided case tour · 3 minutes
New readings arrive.
The last inspection is still unfinished.
A watershed is the land feeding a river or reservoir. After a wildfire, source-water teams must connect changing readings with unfinished field checks. Our agent carries that work forward.
This tour opens with a recorded case. No model call starts from this page.
SHARED WATER
↘≈
schematic watershed
not a geographic map
not a geographic map
Recorded USGS measurements
Simulated field reports
Recorded real agent execution
Follow one case
From signal to field decision
Four moments show what changed, what is known, and where a person still needs to act.
What do we know?
The case ledger preserves the last inspection and its open work.
What happens next?
An agent proposes a focused follow-up for human review.
01 / 04
The readings arrived. A review is due.
Start with the saved source window. The team needs to know what those readings mean alongside work that is still unfinished.
02 / 04
The last inspection left work behind
The earlier report was verified. The inspection was only partly completed. Both facts must carry forward.
03 / 04
A recommendation with its evidence trail
The agent reads the saved evidence and available tools, then proposes a plan. It does not silently close the case.
How was this recommendation made?
The recorded trace will appear here when the case loads.
04 / 04
Now try a human decision
The recorded proposal stays unchanged. Use a separate practice review to see how a human decision is saved and carried forward.
Evidence behind the workflow
Inspect system behavior
Open the evidence behind the workflow.
System map
How the system carries the work forward
Each stage has a job, a boundary, and a visible handoff.
01Collect source observationsScheduled application code retrieves USGS readings.
02Keep the case historyExternal SQLite retains readings, reviews and field records.
03Connect evidence with StrandsThe model selects tools and proposes the next action.
04Run on Bedrock and AgentCoreThe recorded cloud runtime executes the Strands agent.
05Validate and save the proposalApplication checks bind report revisions, sites and permissions.
06Let the field team decidePeople approve, report results and verify evidence.
The case ledger remains external to AgentCore. A human approves the plan before fieldwork moves forward.
Read the verified run record View the full architecture diagram ↗