Section · From the Desk
Writing.
Essays and notes on systems design, AI-augmented workflows, and the tradeoffs that do not make it into release notes. Updated as the draft folder demands.
Why ERP modernization is the real AI opportunity.
The most boring systems in the enterprise are the ones agents are best positioned to change. A brief on why the ERP backbone — SAP, Oracle, the forty-year-old schemas that run purchase orders and payroll — is where GenAI will actually earn its keep, and why the shiny copilot demos will not.
Read the piece →Notes from six months leading GenAI at enterprise scale.
A balance sheet after two quarters embedding AI across internal platforms at Walmart Global Tech. Where it works, where it does not, which models made it into production, and the operating model that kept the whole program honest without starving the rest of the engineering org.
Read the piece →Typed interfaces, honest errors.
A field guide to error propagation that respects the caller: typed failure modes, bounded retries, and a hard rejection of the fallback that silently hides a broken system. The examples are TypeScript and Python; the discipline is language-agnostic.
Read the piece →The middle between a prototype and a product.
Where most AI software goes to die — the territory between a convincing demo and a system people rely on for their jobs — and what it takes to survive the crossing without throwing away everything that made the prototype fast, weird, and worth shipping in the first place.
Read the piece →On the grammar of enterprise software.
A working theory: the difference between a tool people use and one they work around is not the feature list. It is the vocabulary the interface lets them speak — the verbs, the nouns, the shapes of the things they can say. A piece on interface grammar as the load-bearing design decision it actually is.
Read the piece →Hiring for judgement, not tenure.
The resume tells you who has been in the room; an afternoon of hard tradeoffs tells you who belongs there. A short note on running hiring loops that respect both — and on the specific signals that map to staff-plus engineering decisions in the wild.
Read the piece →