A faster, smarter @incident

Last year we launched @incident, our AI agent that’s on-hand in every incident. Most commonly, customers have loved using @incident to draft their updates, explain what’s going on, and find the last time this sort of problem happened.

This week we released a significant upgrade to our chatbot, which applies to both Slack and web interfaces.

Much faster responses

@incident is noticeably snappier now. Especially for simple requests like sending updates and answering questions, responses take 3s rather than 9s!

Smarter answers

It’s also a lot smarter; it’ll be able to balance findings from humans versus AI SRE appropriately, and digest information from various sources into timelines or technical explanations.

There’s a real-life example from a production incident of ours below, when a user asked @incident to explain the current incident.

• Before the upgrade, @incident highlighted technical findings from AI SRE; notably that there was a spike on a telemetry dashboard which could have been related.
• After the upgrade, @incident paid more attention to the confirmed findings from humans in the channel, and pointed the user to exact PR’s that were the cause and fix.

Dialling up @incident’s ability to think also means that it will admit when it doesn’t know something, rather than hallucinating.

Suggests what to do next

@incident now anticipates what the next steps might be, and will end its answer offering to do them for you.

More concise answers

@incident won’t waste time giving you information you didn’t want. Simple questions like “who is the lead?” now receive one-line answers, but questions about technical concepts still go into detail.