Things I've been thinking about: data strategy, building with AI, and what it means to lead in technical spaces.
Large enterprises are not failing at AI because of technology. They are failing because they never made a deliberate organizational decision about how to build together. The result is duplicated agents, incompatible platforms, and capabilities that cannot scale beyond the team that built them.
AI agents are absorbing economic activity at the same moment demographics are shrinking the human tax base. Governments are treating these as separate problems. They are not.
Enterprise data architectures were built for a world where humans did the reasoning and systems moved the data. Agentic AI changes the consumer. The architecture needs to change with it.
I keep getting asked to solve data problems for family and friends running small businesses. They need more than a spreadsheet but can't justify enterprise tooling. So I built a full data lake (raw storage, AI-powered ETL, structured serving, browser-based SQL) on free Google services. Total cost: about five dollars a month.
A large CPG company asked us to help them build an AI strategy. We went to the standard playbooks -- DAMA, TOGAF, Data Mesh. None of them answered the actual question: across our entire data organization, what do we have, what are we missing, and what do we need to build? So we built a framework to answer it.
I built my personal website in one evening. One person. No team. No budget. No months of back and forth. Just AI, a few hours, and every decision made by me. That's remarkable. It's also what keeps me up at night.
Most organizations know they have a customer data problem. What they rarely anticipate is how visible it becomes the moment they try to answer a simple question: who are our customers, and what is the full scope of that relationship? This is what we found when we followed the customer journey end to end, and why the answer required an organizational decision before it could ever require a technical one.
GenAI is not limited by model capability. It is limited by the quality of your business metadata. In a simple SQL agent experiment using cryptic column names and no semantic layer, the only thing that enabled correct reasoning was a governed data dictionary. The agent inspected the schema, read trusted definitions, mapped fields accurately, adapted to execution errors, and returned valid results. Without metadata, it would have guessed. With governed metadata, it reasoned. AI does not replace data governance. It amplifies it.
The customer journey is not just a marketing concept. It is a data problem. Most companies don't have a customer data problem in the abstract. They have it at specific moments: when a prospect becomes a buyer, when a buyer calls support, when a loyal customer churns quietly. At each stage, a different system captures different data under different definitions, with no guarantee the records resolve to the same person. That's the core MDM problem in customer data.