Operations | Monitoring | ITSM | DevOps | Cloud

AI applications can behave unpredictably, potentially leading to errors such as hallucinations or data leaks, even when classic monitoring indicates a successful response. To effectively monitor AI systems, four key areas should be focused on. Implementing these pillars can enhance trust in AI deployments, help manage costs, and identify safety issues before they impact users.

Grafana 13.1 cuts down on GitOps pain by making Git Sync stronger — import dashboards to Git with a click, see your repo READMEs inside Grafana, and sign every commit for security-strict teams.

Learn the two main ways to get data into Grafana Cloud. In this video, we break down how Grafana Cloud connects to over 150 external data sources (like Salesforce, Postgres, and CloudWatch) where your data stays in place, and how you can send raw telemetry into Grafana’s fully managed databases for logs, metrics, traces, and profiles.

This 3-part workshop series covers everything already included in your Sentry subscription and how to actually use it.

At LLMday Lisbon, our Software Engineer, Viktor Vasylkovskyi, highlights the realities of building production AI agents with LangGraph - sometimes getting it right, often learning the hard way. This talk is about what was actually shipped, including a distributed multi-agent setup at PagerDuty. Viktor breaks down the real tradeoffs between LLM-driven and deterministic orchestration, what broke, and how he’d approach it differently now.

Every other noc was just practice for this one.

In this episode, Leon explores some of the new features, functions, updates, and improvements in release 4.18 (from last month) and 4.18.2. For more information, check out these links.

Why do so many AI pilots succeed in testing but fail to reach production? In this webinar, Resolve and IT leaders from RisePoint explore one of the biggest challenges facing enterprise AI adoption today: trust. While organizations are investing heavily in AI agents and automation, many initiatives stall before deployment due to governance concerns, compliance requirements, risk management, and lack of operational visibility.

The real unlock is flagging what actually needs human review, not shipping 5,000-line PRs and hoping for the best.

The era of token maxing is over. When Claude Fable 5 launched last week at $10/$50 per million tokens - double the price of Opus 4.8 - it was a clear reminder that the most powerful model isn't always the right model. Not every task needs the Ferrari. The fastest way to burn your Al budget is sending every request to the most expensive model by default. The real question for the next phase of Al cost management isn't "can this model do the job?" — it's "is it the right model for the job?".