In this episode of Software Engineering Radio, Srujana Merugu, an AI researcher with decades of experience, speaks with host Priyanka Raghavan about building LLM-based applications. The discussion begins by clarifying essential concepts like generative vs. predictive AI, pre-training vs. fine-tuning, and the transformer architecture that powers modern LLMs.

Srujana explains diffusion models and vision transformers, highlighting how multimodal AI is reshaping content creation. The conversation then moves to practical aspects—where LLMs make sense, where they don't, and a decision framework for evaluating use cases. They explore common application patterns such as retrieval-augmented generation (RAG) and agentic architectures, breaking down components like planners, orchestrators, memory, and tools. Key considerations for model selection, evaluation metrics, and safety guardrails are discussed in depth. The episode also touches on prompting strategies, automated prompt optimization, and emerging trends like multi-sensory AI and the "Internet of Senses." Finally, Srujana shares tips on staying current in a fast-moving AI landscape and emphasizes lifelong learning and curated knowledge sources.

Mostly productive day, but wasted WAY too much time trying to update some components on my local machine. Python dependencies are the worst.

[blog] Titans + MIRAS: Helping AI have long-term memory. Is this the approach that comes after Transformers and represents the next major shift in AI models? Google Research introduces these intriguing concepts.

[blog] It’s the People, Stupid. Now, more than ever. The tech is fine, and frankly far ahead of what most of us are ready for. People aren’t free, and you’ve got to make the human costs visible.

[article] 2025: The State of Generative AI in the Enterprise. Some great data here, including looks at which vendors and technologies matter to different departments, and layers of the stack.

[blog] AI Is Forcing Docs To Finally Grow Up. Docs are a product, absolutely. My team treats it that way, and that means having PMs, backlogs, and a customer focus.

[blog] The isolation and loneliness of tech writing may get worse as AI accelerates. Important points from Tom here. I need to think about this more. Remote work and AI could lead to many different types of roles working in isolation.

[blog] Shifting left at enterprise scale: how we manage Cloudflare with Infrastructure as Code. It’s always fun to see the tech choices each company makes. Cloudflare heavily uses Terraform, Open Policy Agent, and a GitLab-based CI/CD pipeline in their IaC setup.

[paper] Cognitive Kernel Networks (CKN): Root Trust for Privilege-Separated Reasoning in High-Dimensional Models. A colleague wrote it, so I wanted to read it. But I didn’t really understand it. Perfect case for AI and NotebookLM which summarized it for me and built a slide deck to explain it better.

[blog] Is your DR plan just wishful thinking? Prove your resilience with chaos engineering. Sure, this was a hot topic a decade ago. But now? It might even matter more as companies have more distributed infrastructure and their businesses are crippled when there’s a tech outage.

[blog] Linux Foundation Announces the Formation of the Agentic AI Foundation (AAIF), Anchored by New Project Contributions Including Model Context Protocol (MCP), goose and AGENTS.md. Along with a handful of others, Google is a founding member of this new foundation. There’s a lot still to be settled when it comes to agent tech, but standards are starting to emerge.

[blog] Gemini CLI Tutorial Series — Part 15: Gemini CLI Extension for Google Workspace. MCP makes a lot of things easier. And this Workspace extension lets you do virtually everything with Gmail, Docs, Chat, and more from the CLI.

[blog] 5 Legacy API Protocols That Refuse to Die. I hope I never write another SOAP web service again in my life. But much of the world still depends on the protocols named here.

Want to get this update sent to you every day? Subscribe to my RSS feed or subscribe via email below:

Our dynamic four-part webinar series, Agentic AI + Copilot Partner Skilling Accelerator, empowers you to harness the Microsoft AI ecosystem to unlock new revenue streams and enhance customer success. Across the four sessions, Microsoft partners can expect to learn how to apply AI tools in no-code, low-code, and pro-code scenarios to build intelligent chat and workflow solutions, extend and customize capabilities, and create advanced, custom AI functionality.

Don't miss the final session in the series, Building Agents with Azure AI Foundry and Azure AI Foundry Agent Service, where you'll learn how to design and deploy intelligent agents with Azure AI Foundry and Azure AI Foundry Agent Service, including multi-agent architectures and key protocols such as A2A and MCP.

The live virtual event is scheduled for December 15, 2025.

Register today to reserve your spot!

Be sure to follow this Partner news blog for all partner related announcements by clicking follow above!