Cursor investigates support tickets across tools
Cursor support engineers report using the agent to investigate code, logs and prior cases, then review a merged escalation and customer reply before sending.
cursor.com
Original prompt
Shared by the source author. See source notes for context.
Create a Linear ticket for the reported bug or user issue.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Kody Fisher and Zach Hudson
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- Cursor support engineers report using the agent to investigate code, logs and prior cases, then review a merged escalation and customer reply before sending.
- Human involvement
- Support engineers review and send the merged output. Reply rules prohibit exposing internal infrastructure, paths, variables and customer-specific information without approval.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Start from the customer symptom and trace relevant local repositories.
- Retrieve time-bounded Datadog traces and identifiers through MCP.
- Search Slack and Notion for current incidents and documented behaviour.
Kody Fisher and Zach Hudson describe how Cursor's own support team combines its local product repositories with MCP access to operational information. An investigation starts with a customer symptom and traces the relevant code, including frontend behaviour and backend policy checks across a workspace containing multiple repositories.
The team brings Datadog logs and traces into the same investigation, searches Slack for exact error strings or request identifiers, and consults Notion runbooks to distinguish a bug from expected behaviour. Their known-issue procedure records whether an issue is known, possibly known or not found, alongside a relevant thread and a next search when evidence is weak. Linear receives the assembled escalation with reproduction steps and trace links.
For repeated work, narrow subagents separately investigate logs, retrieve prior cases, prepare a ticket and draft the reply. Engineers review the merged output and send it. The reply instructions remove internal infrastructure details, environment variables and private error codes, while preserving concrete troubleshooting steps. The authors report widespread internal adoption and higher throughput, but provide no independent measurement. The reusable detail is a shared evidence trail and an explicit review step before customer communication, rather than letting a plausible draft substitute for a completed investigation.
Prompt
Original prompt
Source: Blog · Kody Fisher and Zach Hudson
Frequently asked questions
How are prior incidents found?
The team searches hard identifiers first, then narrows by channel and time, seeking a recent thread with a status, workaround and owner.
What prevents internal details reaching customers?
Reply rules remove service names, infrastructure details, private error codes, paths and environment variables; the engineer reviews the merged result.
Does the source prove a throughput gain?
The authors report higher support throughput from their own team. Their post does not provide an independently reproduced benchmark.
More details
Remaining steps
- Assemble a Linear escalation and a customer-safe reply.
- Have an engineer review the merged investigation before sending.
About this account
Specialist agents gather different evidence sources, refine searches when results are weak and assemble an escalation and reply; human engineers decide what to send.
Firsthand account by Kody Fisher and Zach Hudson; original publication date verified against Blog byline Mar 3, 2026. Outcomes are the author’s reports, not independently reproduced. Original Create Support Ticket command excerpt from the primary account; not the complete command.
Credit: Kody Fisher and Zach Hudson
Source published
Added to Agents In Use
Source checked
Tools and topics
Cursor · Datadog MCP · Slack MCP · Notion MCP · Linear MCP
Technical support · Incident evidence · Customer reply review
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.