Operations | Monitoring | ITSM | DevOps | Cloud

On today's episode, Andrew Hillier chats with Sachin Gharge, Head of Cloud Platform at Scandinavian Airlines (SAS). They discuss AI, agents, Kubernetes, MCP, and optimization. Sachin shares how he and his team are optimizing cloud costs, leveraging automation, and experimenting with agentic AI, including bots and Slack integrations, to make operations easier and more effective for developers and the business.

An Alibaba-affiliated AI agent went full crypto bro. During reinforcement learning, the agent autonomously started mining cryptocurrency, downloaded the tools it needed, and created a reverse SSH tunnel to get around network restrictions. What starts as a funny story about an AI vaping and mining crypto gets a lot more serious when you realize how sophisticated the behavior actually was.

Stop making your AI coding agent guess how your app works with massive prompts. We feed live API traffic and runtime evidence directly into your agent so it can automatically detect bugs, run replays, and fix code fast! Learn more: speedscale.com.

Choosing where your data lives is only half the question. How it's stored matters too. File, object, and block storage are each built for a different job. File storage uses the hierarchy you already know. Files live inside folders, and those folders can live inside other folders. Object storage gets rid of that hierarchy. Each piece of data is stored as its own object, which makes it easy to scale.

What happens when an AI agent can do more than suggest a fix — and actually take action inside your CI pipeline? See Harness Worker Agents in action as an AI agent identifies a failed CI build, determines what went wrong, creates the fix, and gets the pipeline moving toward production again. Worker Agents bring AI-powered reasoning directly into your software delivery pipelines while maintaining the controls enterprises need, including sandboxed execution, scoped credentials, policies, and RBAC.

Jev is a new AI model from Type Safe built around a very different idea: instead of generating long answers, it makes fast, simple decisions. The team reportedly even demoed it playing Doom using nothing but split-second choices. After years of teaching AI models to talk, could the next breakthrough be teaching them to simply decide?