Operations | Monitoring | ITSM | DevOps | Cloud

During emergencies and illnesses, people expect intuitive healthcare services. When multiple tests and reports are involved, patients anticipate that the results will be available to their doctors instantly for quick diagnoses. Waiting for a paper copy of each test result is not feasible.

In the world of large language models (LLMs), two approaches have dominated how we adapt AI to specific use cases: Retrieval-Augmented Generation (RAG) and Fine-Tuning. But the landscape is rapidly evolving with advanced techniques like MoE, LoRA, and GRPO. Let’s explore how these approaches compare and combine to create more powerful AI systems.

Take a drink if you hate that term. Tim O’Reilly has written a post about the huge shift AI is making with our relationship to software engineering (read: The End of Programming as we Know It). Scary title. But the content is less so. Why?

As AI becomes increasingly central to modern software development, the ability to craft effective system prompts has emerged as a crucial skill. Whether you’re building a code generation tool, creating a chatbot, or developing AI-powered features, your success largely depends on how well you can communicate with AI models through prompts. At CircleCI, we’ve spent countless hours working with developers who are integrating AI into their applications.

Not long ago, our co-founder and CTO, Erik Peterson, shared some insights on AI spending. He shared how AI costs currently fall under the write-off-friendly world of R&D. He also acknowledged why DevOps teams might feel it’s too early to start optimizing AI costs. As the saying goes, “Premature optimization is the root of all evil.” But after more than a decade of software development, Erik knows that eventually, research, experimentation, and big ideas need to deliver real returns.