More essays
Agents Got a Harness. Data Needs a Different One.
The harness, not the model, made coding agents reliable. A data harness has to be a different machine, because data has no compiler to catch a wrong answer.
We Were Never Measuring What Users Wanted
Will AI replace data analysts? No. As users state intent in language, the job shifts from measuring what people did to interpreting what they mean.
Data Is Not Software
Anthropic runs self-service analytics on a governed semantic layer, not a chatbot pointed at a warehouse. The same case we have been making.
What OpenAI's Data Agent Teaches About AI-Native BI
OpenAI's internal data agent looks like a chatbot. Inside, it is a context graph. Here is what AI-native BI actually requires, and what most teams will miss.
The Human API Problem
Data teams spend most of their week answering ad-hoc questions instead of building. We call this the Human API problem. Here is what it costs and what to do about it.
The Data-Driven Lie: Why Most Companies Fail at What They Claim to Do Best
Data-driven decision making has become table stakes language. In practice, most leadership meetings still run on whoever argues best. Here is what actually goes wrong, and what would fix it.
Why LLMs Should Never Calculate Your Churn Rate
Most AI-native BI tools let a language model calculate your business metrics. That is a category error. Here is the architecture we built instead, and why.
What is a Decision Context Graph? An Architectural Guide
A Decision Context Graph is the missing layer between your warehouse and your decisions. Here is what it is, how we build one in four hours, and why it matters now.