Blog
Thoughts on engineering, leadership, and product craft.
Latest Articles
First Principles Thinking: You Can Borrow an Answer, Not the Reason
Most engineering models were absorbed rather than built, so when the obvious lever stops working, there is nothing underneath to check. First principles thinking is the fix, but the usual account of it stops at decomposition, which is how careful reasoning on a bad premise produces a confident wrong answer. Two more steps close it: predict something specific, then confirm it from a direction that shares no assumptions.
The Ambiguity Ladder: Why the Next Level Is a Different Job, Not More of the Current One
Most engineers stall because they treat leveling like a progress bar: ship more, ship faster, and eventually tip over into the next level. But each rung is a different job, defined by how much ambiguity you are trusted to own, and it rests on a knowledge base that quietly depreciates. Here is how to find the rung you are actually on, whatever your company happens to call it.
The Patterns Hiding in Your AI Workflow
LLMs are where web design was in the late 1990s: creative chaos before the patterns solidified. This article maps the workflow patterns already emerging from daily practice: from prompt layering and context shaping to multi-pass pipelines and the shifting boundary between tool and collaborator.
Performance Tests That Produce Numbers, Not Answers
Performance tests often run successfully while answering no clear question. This article explores why performance testing fails when intent is missing, and how a lightweight, experimental mindset can turn metrics into meaningful answers.
Running LLMs Locally: Overconditioning, Right-Sizing, and Multi-Pass Pipelines
A practical look at running LLMs as system components rather than assistants. Based on processing thousands of ASR transcripts, this article covers prompt overconditioning, model right-sizing, local execution tradeoffs, and why multi-pass pipelines with validation outperform single, all-in-one prompts.