Operations | Monitoring | ITSM | DevOps | Cloud

Note: A version of this post originally appeared on the CSS Electronics blog. Martin Falch, co-owner and head of sales and marketing at CSS Electronics, is an expert on CAN bus data. Martin works closely with end users, typically OEM engineers, across diverse industries, including automotive, maritime, and industrial. He is passionate about data visualization and AI—and he’s been working extensively with Grafana Assistant.

AI should speed up innovation, not inflate your cloud bill. But today, the biggest GenAI challenge for SaaS teams isn’t model quality; it’s cost. And increasingly, that cost comes from AI cost sprawl. That’s not because anyone is doing something wrong, but because AI operates differently from the cloud services we’ve all spent a decade learning how to manage.

Since launching Harness in 2017, we’ve been on a mission to unlock faster innovation by removing the bottlenecks that slow software engineering teams down. From day one, we believed that the biggest obstacles in engineering weren’t in writing code — they were in everything that followed.

Most AI programs don’t fail on model quality. They fail because the experience makes people either over-trust or quietly avoid the system. Employees often use AI more than leaders realize, frequently without training or guardrails. Interfaces that just “show an answer” without confidence, provenance, or recourse create two risks: blind reliance and shadow use.

If there’s one thing clear from our AI-powered observability webinar, it’s that observability has officially graduated from a “nice-to-have” to a business-critical discipline, and AI is helping lead that charge. Our webinar brought together guest speaker Stephen Elliott, Group VP at IDC, and Ranbir Chawla, former SVP of Engineering at RB Global, for an hour of insights that mixed data, experience, and hard-won lessons from the trenches.

You've set up the Cortex MCP and connected it to your AI assistant and IDE. You ask about service ownership, check a Scorecard or two, and it works. You're impressed by how much faster this is than clicking through the web UI. Now you're wondering what else you can do with it. I'm willing to bet we've hit a nerve with that "hypothetical" scenario. The Cortex MCP works exactly as designed, but it's deceptively difficult to know which questions to ask and when to ask them.

We launched agent monitoring earlier this year, allowing our users to instrument LLM usage and tool calls in their applications. However, we only had Agent Monitoring support for Python and JavaScript. We’ve been working on creating an Agent Monitoring SDK for.NET — specifically for Microsoft.Extensions.AI.Abstractions.

For our last episode of 2025, we’re focusing on Datadog releases announced at AWS re:Invent. Join Jeremy to see how you can manage logs at petabyte scale in your infrastructure, eliminate unneeded costs in Amazon S3 buckets, build agentic workflows, and detect credential leaks. Later in the episode, Scott spotlights how you can connect your AI agents to Datadog tools and context with our MCP Server.