What Doom taught us about AI-assisted incident response

sylvainkalache1 pts0 comments

What Doom taught us about AI-assisted incident response | Rootly

Download  PNG<br>Light

Dark

Download  SVG<br>Light

Dark

Download all assets

Switch from PagerDuty

Log in

Book a demo

Product

Product

On-Call

Incident Response

AI SRE

Retrospectives

Status Page

Platform

Catalog

Integrations

On-Call Health

Changelog

Jul 16, 2026

The on-call widget, 24-hour time, and a new Alerts table.

Solutions

Size

Enterprise

Startup

Compare us

PagerDuty

Opsgenie

JSM

Teams

Engineering

Security

How Canary built incident management from scratch on Rootly while its engineering team quadrupled.

Replit brings structure, speed, and better decision-making to incident response with Rootly.

How Caribou saves 200+ engineering hours a year with Rootly.

Why Capa replaced incident.io with Rootly.

How Reactiv outgrew incident.io and moved to Rootly for real AI and on-call without the upsell.

How SpotDraft replaced manual follow-ups with synced, measurable incident management on Rootly.

How GRAIL replaced a manual, ad hoc incident process with Rootly and cut manual effort by 80%.

How Cora cut time-to-acknowledge on critical alerts to

Why CTSO Central chose Rootly over incident.io.

Poll Everywhere switched from PagerDuty, cut on-call tooling costs and made response faster with Rootly.

Why Momentum didn't choose FireHydrant, and how they minimize the number of engineers on-call with Rootly.

KnowBe4 switched from PagerDuty and turned Incident Response into a repeatable, predictable system with Rootly.

Achievers switched from PagerDuty and productized reliability on Microsoft Teams with Rootly.

How Motive migrated from Opsgenie and achieved 99.99% reliability with Rootly.

Clay switched off Opsgenie to use Rootly to streamline incident management.

Webflow shutdown PagerDuty to use Rootly to build a proactive reliability culture.

How Wealthsimple uses Rootly to create a culture of wellness and psychological safety for incident commanders.

How Upstart uses Rootly to create an incident response system that grows with them.

How ROLLER scales globally while keeping a lean SRE team.

How Lucidworks uses Rootly to create bespoke incident management for their distinct product offering.

Customers

Resources

Engineering

Documentation

API Reference

System Status

Help & Support

Guides

AI SRE & Automated RCA

On-call Operations

Incident Response

Incident Communications

Blameless Retrospectives

Resources

Blog

Humans of Reliability

Rootly AI Labs

Rootly Academy

Partners

Latest Blog<br>Jul 16, 2026

What Doom taught us about AI-assisted incident response

Pricing

Log in

Book a demo

Switch from PagerDuty

Back to blog

What Doom taught us about AI-assisted incident response<br>Sylvain Kalache<br>July 16, 2026

Introducing Doom Agent Arena, an open-source real-time game environment benchmark where AI agents control Doom players via MCP, and fight each other across multiple rounds.<br>By building Doom Agent Arena, our goal was to expand our research scope to better understand how LLMs can assist incident responders in resolving outages as quickly and effectively as possible. Although built in a game setting, the benchmark targets reliability-relevant LLM skills: interpreting dynamic environments, reasoning about consequences, recovering from failed plans, and adapting under time or cost constraints.<br>Let’s dive into how we built it, the methodology, and the findings.

How Doom Agent Arena was built<br>Most other Doom benchmarks feed game frames to a vision model. But because our goal was to test LLMs' reasoning capabilities, our benchmark is instead letting agents observe the Doom game state via our MCP server as structured JSON, including the map and resources, and submit high-level plans for Doom to execute.<br>Latency was one of the main engineering challenges. Doom is built for humans, who make decisions in milliseconds, but a model round-trip takes far longer, so we had to balance the two. The tool calls themselves are fast, but the model's thinking takes up to 8 seconds per decision. That ruled out letting the model play in real time. So we went with a mixed approach: the model makes the high-level decisions, like what the player should do and where it should go, while Doom's engine handles the execution (moving, aiming, shooting).<br>Methodology<br>The game is a one-on-one Doom deathmatch on a walled arena stocked with resources: health packs and a shotgun. Two AI agents each control a player, and the goal is simple: kill the opponent before they kill you. Agents have to read the map, find routes around the walls, grab the shotgun and health at the right moments, and decide when to push a fight and when to retreat to heal.

To keep it fair, we built a custom symmetrical map so neither side starts with an edge. For the findings we will present, we ran four of OpenAI's models against each other across 120 matches, 20 rounds per pairing, swapping spawns at the halfway mark.<br>Findings<br>Model gpt-5.5 finished first at a 66%...

rootly incident doom response from pagerduty

Related Articles