The task
When something breaks, an operator or the monitoring files a ticket.
Before a fix, an engineer has to read the ticket and collect logs, metrics and data to find where it fails.
Technical support
A first look at every incident ticket
The agent picks up each new incident ticket, gathers the facts and posts the likely cause to the support chat.
OPS-4182 · Checkout errorsOn new tickets
OPS-4182: Some users see “Something went wrong” on the checkout page since about 13:40.
Tickets Read OPS-4182
Logs Search errors since 13:30 · 1,240 lines
Metrics Check the error rate · 4% of requests
Cluster Check services · checkout-api restarting
Team chat Post to the support chat
checkout-api has been running out of memory since the 13:35 release and restarts every few minutes. Most likely cause: the new image cache. I posted the details to the support chat. Rolling back to the previous version should stop the errors.
Go-ahead needed1 of 1
Cluster · Roll back service
- service
- checkout-api
- to version
- previous (13:10)
- ticket
- OPS-4182
Message…
AgentSupport Engineer
ModelGPT 6.1 Sol
ReasoningAuto
Screens show sample data. The case is real use in the group that built Sigma.
How the agent does it
- A small service watches the ticket queue. When a new ticket appears, it starts the agent.
- The agent reads the ticket. If the operator’s note is unclear, it asks them in the ticket.
- It checks the data, metrics, logs and services.
- It posts its findings to the support chat: the logs, the problem and the likely source of the error.
- It can fix simple things, like restarting a service, but only with a person’s go-ahead.
What changed
- When an engineer opens the ticket, the facts and a first guess at the cause are already there.
In numbers
Reach
9support engineers worked with the agent
An earlier 30-day period
Frequency
53conversations with the agent in 30 days
An earlier 30-day period
Finishedcompany-wide
88%of agent runs across the company are completed
Last 30 days
Time
~1.5 minper agent run (median)
An earlier 30-day period
Production data inside the group, read-only. A run is finished when it ends with the status “completed”.
More from our own use
Finance, sales, product and operations
Answers from company data, without waiting for an analyst
People ask about the company’s data in plain words. The agent writes the queries and answers with a table.
Read the case Client teams, operations and planningReports that build themselves on a schedule
Agents build daily and weekly reports on their own and post them where the team works.
Read the case Customer supportFirst line of customer support
In one of the group’s products, the agent takes new customer requests, looks up the account and brings in the teams it needs.
Read the case