Case studies / Ops agents
Ops agents
Agents that watch production: alerts, incidents, failure diagnosis and security work.
Approach
26 write-ups
Tag:
-
MongoDBNewOps agents
How Observe.AI Used MongoDB Resharding to Scale Its AI Platform
Problem The platform needed more database capacity without downtime as data grew rapidly.
Solution They used MongoDB Atlas resharding to redistribute large collections across additional shards, keeping the service online while expanding storage.
Read on mongodb.comDetails -
DatadogNewOps agentsInvestigate Kubernetes health and processes with the Datadog MCP Server
Problem Engineers struggled to quickly find and interpret Kubernetes health and process data during incidents.
Solution Datadog added Kubernetes health tools, Live Processes search, and a processes skill to its MCP Server, letting AI agents retrieve metrics and process info in one call.
Read on datadoghq.comDetails -
Grafana LabsNewOps agents
Run cost-effective, always-on agents to monitor your environment: the engineering behind Watchers
Problem Alert rules miss unpredictable failures and running many agents can become too expensive.
Solution Grafana built always‑on Watchers that learn normal system behavior from data sources, trigger alerts on drift, and stay cheap by optimizing runtime and resource use.
Read on grafana.comDetails -
Grafana LabsNewOps agents
Any agent anywhere: How to use Grafana Cloud with Claude, Cursor, and any other AI agent
Problem AI agents couldn't see system observability data, limiting troubleshooting.
Solution Grafana built an open agentic observability platform that lets any AI agent access Grafana Cloud metrics, logs, and traces via APIs, enabling context‑rich investigations.
Read on grafana.comDetails -
NVIDIANewOps agents
How NVIDIA Uses AI Agents to Streamline AI Factory Operations
Problem SREs had to check many dashboards manually to find why training performance dropped.
Solution NVIDIA built an AI Factory Operations Agent that uses multiple specialized agents, connected via a protocol, to diagnose issues and suggest fixes through a chat interface.
Read on developer.nvidia.comDetails -
NVIDIANewOps agents
How NVIDIA Factory Operations Blueprint Coordinates AI Agents to Investigate Production Issues
Problem Unplanned factory downtime costs millions and investigating issues takes too much time.
Solution Created a factory manager AI agent that connects to existing data and tools, monitors operations, finds root causes, and directs robots to fix problems.
Cost savings up to 20%
Read on developer.nvidia.comDetails -
HoneycombNewOps agents
Honeycomb Team Agents: AI SRE Done Right
Problem SRE teams spend too much time reacting to alerts instead of preventing problems.
Solution Created AI agents that read code, logs, runbooks and tickets, set goals, run continuously, and proactively suggest fixes or actions.
Read on honeycomb.ioDetails -
VercelNewOps agents
Vercel Agent can now validate alerts before paging
Problem Too many non‑critical alerts were paging on‑call engineers, causing noise and fatigue.
Solution They added an Agent that reviews each monitor alert, checks logs and metrics, and only forwards alerts classified as Critical to the paging webhook.
Read on vercel.comDetails -
PostHogOps agentsEngineers are shipping faster, so we made our CI self-driving to keep up
Problem CI maintenance couldn't keep up with the surge in pull requests.
Solution They built internal engineering analytics and self‑driving agents that watch CI data, detect problems, and automatically open or fix PRs.
11% of merged PRs were self‑driving
Read on posthog.comDetails -
NetflixOps agents
Making Failure Diagnosis Reliable with Pensive AI
Problem Half of failed jobs were unclassified, wasting engineer time.
Solution They added an LLM layer to read logs, extract error details, and suggest fixes, while keeping existing rule‑based logic for known cases.
AI extraction covers 100% of failures
Read on netflixtechblog.comDetails -
monday.comOps agentsHow We Gave AI Agents the Keys to Production and Still Sleep at Night
Problem Infra engineers were overwhelmed by repetitive, judgment‑heavy tasks on production infrastructure.
Solution Created a platform where LLM agents use guarded tool calls to read, propose, and approve changes to Terraform, Helm, and GitOps, operating within shared safety rules.
Read on engineering.monday.comDetails -
ReplitOps agents
Replit Monitoring Agent: Your very own software engineering team
Problem Small teams must manually find and fix app outages, which wastes time and frustrates users.
Solution Created an autonomous monitoring agent that watches app health, diagnoses problems from logs, writes and validates a fix, and can deploy it automatically.
Read on replit.comDetails -
SourcegraphOps agents
php2shell: patching PHP is easy, finding it is the hard part
Problem Teams struggled to locate every PHP runtime in their code to apply urgent security patches.
Solution Sourcegraph built Deep Search to scan repositories for PHP binaries and used an agentic batch‑change system to automatically update all found instances.
Read on sourcegraph.comDetails -
AmazonOps agents
Graph-centric agentic intelligence
Problem Ops teams needed AI agents that could understand complex network graphs to pinpoint failures fast.
Solution Created an agent that walks real‑time dependency graphs, runs cascaded analytics, and uses causal inference to locate root causes automatically.
95% fault localization accuracy in under 30 seconds
Read on amazon.scienceDetails -
AirbnbOps agents
Beyond the model: Engineering AI infra with scientific judgement
Problem Need a repeatable, auditable way to analyze large support chats for AI product safety.
Solution Built an investigation framework that extracts, embeds, and clusters data, automating the methodology so teams can run analyses at scale without manual curation.
Read on medium.comDetails -
HoneycombOps agents
How Canvas Powers the AI Agent Development Feedback Loop
Problem Users struggled to track agent performance and fix issues across scattered tools.
Solution Built Canvas to bring telemetry, code history, tickets, and chat into one workspace, letting teams investigate, prioritize, and validate changes automatically.
Read on honeycomb.ioDetails -
RampOps agentsHow we built OCA, our AI on-call assistant
Problem Incidents took too long to diagnose and resolve, slowing down engineering teams.
Solution Ramp created OCA, an AI on‑call assistant that scans alerts, investigates root causes, suggests actions, posts updates to Slack, answers follow‑up queries, and even drafts code fixes.
Read on builders.ramp.comDetails -
ShopifyOps agents
How River takes security work from a fix to merge
Problem Security teams struggled with slow, manual patch handling and a growing backlog of vulnerability fixes.
Solution They built River, an AI agent in Slack that automates checking, updating, and merging patches, and routes decisions to engineers when needed.
Backlog down 70%, merge rate rose to 80%
Read on shopify.engineeringDetails -
monday.comOps agents
I built a monitor. Nearly 200 engineers turned it into a colleague.
Problem Engineers spent too much time on repetitive on‑call incident triage.
Solution Created an on‑call assistant that automatically checks metrics, logs and error spikes, runs scheduled jobs, and suggests actions, letting the team evaluate its impact.
Read on engineering.monday.comDetails -
ReplitOps agents
The Self-Driving Company
Problem Teams needed help with repetitive tasks like incident triage and code reviews, slowing progress.
Solution Replit built a network of AI agents that take goals, gather context, perform work, verify results, and hand off to humans when needed, automating tasks across the company.
Read on replit.comDetails -
HoneycombOps agents
Multi-Agent Collaboration on a Shared Canvas
Problem Engineers needed to debug production issues together without their AI assistants interfering with each other's investigations.
Solution They created Honeycomb Canvas, a shared whiteboard where each engineer has an independent AI agent that can run in its own isolated micro‑VM yet see peers' work, enabling coordinated, parallel debugging.
Read on honeycomb.ioDetails -
HoneycombOps agents
What Customers Are Doing With AI and Honeycomb
Problem Engineers struggled to keep up with faster code releases and needed quicker incident triage.
Solution Created an AI‑driven incident‑triage bot that pulls Honeycomb traces, logs, and Kubernetes data to give on‑call engineers full context automatically.
Read on honeycomb.ioDetails -
RampOps agents
Agentic Risk Operations
Problem Risk teams spent too much time manually handling alerts and investigations.
Solution Ramp created an LLM‑powered agent that automatically retrieves data, runs checks, and takes remediation actions to streamline risk operations.
Read on builders.ramp.comDetails -
Hugging FaceOps agents
Holo3.1: Fast & Local Computer Use Agents
Problem Users needed computer-use agents that work on desktop, mobile, and any agent framework, not just cloud.
Solution Released Holo3.1 with quantized checkpoints (FP8, Q4 GGUF, NVFP4), function‑calling support, and smaller model sizes, boosting robustness and performance across environments.
25% faster, accuracy up to 79.3% on mobile
Read on huggingface.coDetails -
ThumbtackOps agents
Leveraging AI to build a faster iOS CI system
Problem iOS pull requests waited 30‑40 minutes for CI, slowing development.
Solution They built Smart Test Selection, which uses runtime coverage to run only tests affected by code changes, cutting CI time by about half.
CI time reduced ~50%
Read on medium.comDetails -
SlackOps agents
Streamlining Security Investigations with Agents
Problem Security engineers were overwhelmed reviewing billions of alerts each day.
Solution They built AI agents that automatically pull data, run queries, and suggest actions to investigate alerts, using tool‑calling and multi‑step reasoning.
Read on slack.engineeringDetails
No write-ups match these filters.