Thank You, Goodbye and Good Luck!

So long, and thanks for all the fish - a personal update

Photo by Ricardo Rocha on Unsplash

Today marks my last day at Cube.

Firstly, I’d like to say thank you to everyone I’ve worked with there and who ultimately provided Michael and I with an exit where we were able to make all our investors whole and to continue to work on applying AI to semantic layers.

I’m lucky enough to be in a position that I am able to make ends meet from my other income. So please don’t take this as one of those posts eliciting sympathy for a difficult situation post-job, this is not the case - I’m just reflecting.

I still believe that Cube has the best universal semantic layer out there, and it is the only fully functional open source one of enough significance to be considered an open standard. Godspeed to the Cube team as they continue on their agentic analytics journey. I truly wish them the best and every success on this journey.

I still believe that semantic layers are important and even more so than when I first started writing about them a few years ago. I still remember the magic of using LookML for the first time in 2019. It’s been a bumpy road for semantic layers in that time, but now that the impetus from agentic analytics is driving our industry forward, this time by stakeholders dragging us forward instead of technologists, I think we’re nearly there. Every cloud data platform provider now except AWS… Databricks, Snowflake, GCP, MSFabric, now have a semantic layer of some form that isn’t just a knowledge graph but also has a compiler.

The question I have for the future is whether the semantic layer needs to be seen or can fade into the background. Semantic layers need to exist and need to be codified and compiled, but does that mean humans need to see it or touch it directly? Can AI maintain semantic layers in an invisible way for normal operation? Sure, if you ask the definition of a metric or how the data model fits together, it can answer you. This doesn’t have to mean that human engineers maintain it as code. This is the way that many BI tools that have semantic layers are going. To start with, they are offering extension and maintenance using AI, then they will offer initial creation, then agents will simply show you the code and explain changes they will make… will they always have to show you the code?

It will feel like the formation of collective neural pathways to use data. The pathways are reused, the pathways provide fast and consistent operation, the pathways slowly evolve as definitions and models change over time. An automatically managed codification of collective truth with data. I’m excited to see who succeeds here first!