Operations | Monitoring | ITSM | DevOps | Cloud

Auditable Agent Judgment: Jev's Decision Primitive at Harness Agents do more than generate - they continuously make judgments between reasoning and action. This article explores how typed, probabilistic decisions can become a first-class primitive for building agent loops that can be evaluated, governed, and trusted.

Alert routing often starts simple. A team creates a few contact points, adds some label matchers, and builds a notification policy tree that sends each alert to the right destination. But alerting configurations rarely stay simple. As an organization grows, its notification policy tree must accommodate more teams, services, and routing requirements. Changes for one team still require editing a global configuration, making ownership less clear and independent provisioning harder.

You're producing more trace data than you want to pay to store, so you sample. A fixed 1-in-100 rate cuts your bill, but it's blind. It keeps 1% of your errors, 1% of the requests to that rarely-hit route, and 1% of the health checks, all at the same rate. The noisy traffic you care about least dominates what you keep while the traces you need during an incident are the ones most likely to be gone.

On September 4, Rapid7 Labs published research on a Linux espionage toolkit found at two organizations in South Korea. The centerpiece is a backdoor the researchers call "Ted," which was hidden inside a modified HAProxy build running on the victims' load balancers. Rapid7 attributes the campaign to North Korean state-sponsored actors with medium confidence. The same toolkit tampered with multiple tools across the victims' systems, including an SSH keylogger.

Helping federal agencies reduce digital friction, strengthen mission continuity, and make better use of limited resources.

Why do the same complaints about call quality and slow file transfers keep coming back after the network has been checked and declared healthy? Packet loss is usually the answer, and it survives investigation because it degrades the services people use without taking anything offline. That combination makes it expensive. Equipment gets rebooted, cables get replaced, and tickets go to the internet service provider, often with nobody knowing which segment of the path is discarding traffic.

Most networks were never designed. They were extended, one switch and one VLAN at a time, until a single core failure took the site down and nobody could find the diagram. According to the Uptime Institute Annual Outage Analysis 2026, 57 percent of organizations said their most recent major outage cost more than $100,000. Network design is how you stop paying that bill. You decide the network topology, the addressing and the hardware on purpose, before the cabling goes in.

When an urgent situation occurs, organizations need more than a way to send a message. They need a communication strategy that considers who needs to receive the message, whether they need to take action and how quickly they need to respond. This is where the difference between on-call alerting and mass notification becomes important. A critical IT incident, for example, may require an immediate response from a specific on-call engineer or incident response team.

On August 27, Google sent us an email. We did not. We barely recognized the hostname. None of us had thought about dev-docs.trackjs.com in about ten years.

During a recent Agent Hackweek, an internal Sentry event that gives us a week to build any AI or agent project we want, a colleague pitched me on writing the Laravel AI integration. The goal was to give agents built with Laravel AI the same Agent Tracing support we already have for other frameworks. I liked the idea, he built Sentry’s Agent Tracing for Python based agents before which meant he already had domain knowledge.