It has been just over three years since ChatGPT was first released, and large language models (LLMs) captured the world’s attention. The idea of an all-knowing computer that operates in natural language holds the promise of changing how we live and how businesses operate. Unsurprisingly, like HAL 9000 in 2001: A Space Odyssey, LLMs have produced results that aren’t always what we wanted.
In 2026, the conversation about generative AI is shifting from building models to having those models operate within business applications and as autonomous agents. Many enterprise organizations spent 2024 trying to understand what was becoming possible and not get left behind as their competitors raced to adopt AI. The projects started in 2024 often delivered great demonstrations of a single use case, with the AI very visible as a chatbot providing human-like assistance. As 2025 rolled around, those same projects struggled to expand to broader use cases. We started hearing about AI projects stalling and studies suggesting AI wasn’t delivering the promised business value. Two familiar themes emerge from these studies: skills shortages and inadequate infrastructure. It seems to me that these are two of the top issues with every new enterprise IT trend, from DevOps, through cloud, virtualization, and even client-server applications.
At Tech Field Day, we are kicking off 2026 with AI Infrastructure Field Day 4, where we will hear from vendors offering solutions to your AI infrastructure challenges. We have an awesome panel of delegates who will ask those vendors questions about the real-world of infrastructure and seek answers on what AI will mean for building and operating that infrastructure. I expect the critical questions will be around preparing our infrastructure for AI-enabled applications. Each AI Infrastructure Field Day event has a unique set of perspectives from the combination of presenting companies and our invited delegates.
In September 2025, at AI Infrastructure Field Day 3, we saw a focus on building AI platforms, with discussion about whether enterprises should build and maintain their own platforms using open-source tools or use a vendor platform to deliver rapid value without risking mistakes in a custom deployment. Our January 2026 event will focus more on building and operating the network infrastructure required for AI training or inference. I expect to see more discussion of the different requirements and solutions that suit inference versus training and fine-tuning. Few enterprise organizations will build their own foundation models; most will finetune existing models and augment them with RAG solutions. Inference involves a highly distributed workload and long-term operation, where cost optimization is vital.
Another AI project challenge is the “old garbage in, garbage out” problem, where the data in our business is poorly structured and poorly labeled, hindering its reuse by AI applications. We want to couple our corporate knowledge to our AI applications, often using RAG, yet we don’t really understand what corporate knowledge we have. Sometimes it is easy; recognizing PII in source documents is something where AI excels. Adding guardrails to prevent the AI-enabled application from leaking this data should be our best practice. Other times, it is more difficult; ensuring an AI application respects the permissions of the human asking for information can be difficult if we never tracked that human’s access before the AI.
Join me for the livestream of AI Infrastructure Field Day on Wednesday, January 28th, at 8 a.m. Pacific Time and again on Thursday and Friday at the same time. You can find the stream on the Tech Field Day website and across the Techstrong family of sites. The best place to interact live with the delegates and sponsors is on the Tech Field Day LinkedIn page.
