Investigate any issue, from anywhere.
This use case shows how with Sagy, we can create an agent that runs a complete investigation workflow: understanding the issue, gathering context, checking relevant code and database records, and surfacing existing tickets.
Inputs
Tools touched
Sagy turns inbound issue reports into structured investigations. The agent reads the report, pulls context from your tools in parallel, and surfaces the root cause and next action. Before anyone opens a tab.
Use a configurable Sagy agent to investigate a customer issue from Slack (or email, or a call), determine whether it is part of a known incident, and surface the appropriate next action. The goal is to reduce manual investigation time by using the agent's workflow, connected tools, and feedback loop.
- Works from any inbound channel: Slack, email, or call transcripts
- Cross-checks Jira, GitHub, and your database in parallel
- Tells you if it's user-specific or part of a known incident
- Workflow stays auditable. Every step and source is logged.
From inbound message to root cause.
Six steps, fully auditable. Each step links to the moment in the walkthrough video.
- 1Watch at 0:34
Receive and review the customer issue
- Open the Slack message (or forwarded email / call transcript) from the customer.
- Read the issue carefully and identify the core symptom (for example: "cannot log in").
- Confirm the request is suitable for automated investigation before proceeding.
- Capture any key identifiers mentioned, such as the user email or account ID.
- 2Watch at 1:20
Trigger the Sagy agent to begin investigation
- In the Slack thread, invoke the Sagy agent using the configured command or prompt.
- Provide a clear instruction such as:
Please investigate. - Include the relevant context from the issue so the agent starts with the correct request.
- Submit the request and wait for the agent to begin processing.
- 3Watch at 1:51
Ensure the agent has the right workflow and tool access
- Verify that the Sagy agent is configured to execute the required workflow for this type of issue.
- Confirm the agent has access to the necessary knowledge sources and tools.
- For an investigation, ensure access to Jira tickets, GitHub code, and database records.
- Make sure the workflow instructions are aligned with the investigation process.
- 4Watch at 3:21
Let the agent analyze the issue and identify the root cause
- Review the agent's response once it completes the investigation.
- Check whether it identifies the issue as a user-specific problem or a broader platform incident.
- Use the agent's conclusion to determine the next action.
- If the agent finds an existing incident, confirm the incident details and status.
- 5Watch at 4:08
Validate the workflow execution and tools used
- Open the workflow details in the Sagy platform.
- Review the request, the agent used, and the output workflow.
- Confirm which tools were used during the investigation: Database, Jira, GitHub.
- Review the workflow description and instructions to ensure they match the intended process.
- 6Watch at 4:34
Review performance and update the workflow with feedback
- Compare the time spent by the AI agent versus the manual process.
- Record the time saved for reporting or process improvement.
- Provide feedback on the agent's output if adjustments are needed.
- Confirm that the workflow updates automatically after feedback is submitted.
Cautionary notes
What to watch for
- Do not assume every issue is user-specific. Verify whether it is part of a broader incident.
- Ensure the agent only has access to approved tools and data sources.
- Review the agent's conclusion before taking customer-facing action.
- If the workflow is misconfigured, the agent may investigate the wrong systems or miss important context.
Tips for efficiency
Get the most out of it
- Include the user email or other identifiers in the initial request to speed up investigation.
- Keep workflow instructions concise and specific so the agent can follow them consistently.
- Use the agent for common, repeatable issues where the investigation path is predictable.
- Collect feedback after each run to improve future executions automatically.