The thinking spine
Thinking
Essays and field notes — the public version of things I figured out the hard way, written so they can be applied again.
8 results
The Architecture You Remove Still Has to Exist
Teams often celebrate removing components, abstractions, or processes in the name of simplicity. The harder question is where the responsibility moved. Complexity rarely disappears. It usually changes ownership.
The Rules You Never Wrote Are Still Running Your System
Teams often believe they are debating architecture when they are really discovering undocumented business rules. Making those rules explicit is one of the highest leverage activities in building systems that people can understand, evolve, and trust.
The Most Useful AI in My Week Just Explains Itself
The AI that's saved me the most time this year doesn't automate anything. It assists, and then it shows its reasoning — and that second part is the whole game.
A Pipeline That Fails Loudly Is a Feature
The instinct to make pipelines 'resilient' by continuing through errors usually makes them more dangerous, not less. A short note on choosing loud failure on purpose.
The Customer Is Always Somewhere in the Data Model
It's easy to treat a data model as a purely technical artifact. But every table, every default, every simplification quietly distributes benefit and friction among real people. Naming that distribution is part of the job.
Architecture Is What Still Works After the Diagram Meets Reality
A clean diagram is a hypothesis about a system that doesn't exist yet under conditions that haven't arrived. Real architecture is defined by how the design behaves under the stress vectors you know are coming.
BI Is Decision Infrastructure, Not Decoration
Most organizations have more dashboards than decisions. Treating BI as infrastructure — with the reliability standards that word implies — is what separates a tool people depend on from a gallery nobody visits.
Data Platforms Are Trust Systems Before They Are Technology Systems
We talk about data platforms as if the problem were technical — pipelines, warehouses, latency. The real problem is human: will anyone trust the number enough to act on it? That changes what you build first.