will warnock

Big Data London 2026

I was fortunate to spend a day at Big Data London 2026. This conference was all about data, analytics and AI. This whole industry is up in the air at the moment so I was quite excited to get some exposure to the innovation that’s going on. The likes of Google, Monzo, Snowflake and even Microslop were all in attendance as exhibitors, in addition to what felt like 100 other companies. I got the most out of my day by sweeping the floor, walking up to booths and talking to the people showcasing their products. I enjoyed getting pitched on what the product did, what challenges it tried to solve, and getting swiftly referred to a “technical expert” whenever I asked a curly question. In all fairness, I did plan on picking some talks to sit down and listen to, but the first one felt more like watching a 20 minute ad, so I didn’t hang around to watch another. Now the takeaways.

semantic layer semantic layer semantic layer semantic layer semantic layer semantic layer semantic layer semantic layer semantic layer

The term “semantic layer” has reached peak hype with most exhibitors mentioning it as part of their product or pitch. For those unaware, the semantic layer is where the mapping between business terms and data rules are stored (e.g. “an active user” = “any user that logged into the platform in the past 30 days”). This is useful because it makes LLMs respond to data questions more consistently. My take is that this is all well and good, but paying for your “AI analytics platform with a semantic layer” doesn’t simply solve this knowledge gap overnight. Semantic layers are tribal knowledge in many companies and rounds of discussions and consulting would surface what needs including in the semantic layer, not just adding another SaaS tool to the tech stack.

Next up is that agents seem to be everything. We’ve already quickly moved on from chatbots that return an answer in text. The next evolution of this is agents that can do things with that information. Creating visualisations, sending Slack messages or evaluating themselves for quality. My guess as to why no one was demo’ing a chatbot is because the AI models + harnesses can do all of that without needing another product. It looks like these data analytics companies are trying to move into offering different value (like governance) than going feature-for-feature with the likes of Claude Code, Codex, or the agentic harness du jour.

One company reinvented the wheel created an abstracted language for SQL that its AI agents could write in. They told me this language is then compiled into SQL deterministically so that, in their words “the AI never writes the SQL”. My reaction to this was well yeah, but the agent wrote the query probabilistically into your voodoo SQL language, so all you’ve done is move the problem 🫠.

While a bit of this post is taking jabs at some of the things I saw, I am really excited to see how this all shakes out. Everyone can see the utility of agents working with data, but the best way of doing it hasn’t been figured out. The “Steve Jobs” moment is yet to come where someone wraps this whole concept up in a bow and delivers an elegant final product that both the expert and common man can use. As of the time of writing this, I’ll be sticking with my folder of markdown files that provide context for my handful of agents. Old-school, maybe, but I’ve got significant value from my “never use an em-dash anywhere in this project” rule. In the meantime, I’ll be closely following the innovation in this space as new best practices emerge.