Long-form
What Happens When We Let AI Make Decisions
In July 2026, OpenAI and Hugging Face disclosed an incident in which OpenAI models exploited security vulnerabilities while pursuing an evaluation goal.
Writing
Here is what I was trying to figure out. Here is what I found.
Technical writing on building technology, AI agent patterns, the economics of software, and the work of senior technical leadership. Long-form pieces publish when ready. Field notes keep the cadence warm.
Long-form
Why AI Hasn't Sped Up Your Software
or...we are running really fast, how come we aren't going anyehere.
Long-form
Three Lenses and a Translator
Part 3 of the changing role of data analysis with LLMs
Long-form
Don’t Put an LLM in the Analyst’s Seat
Part 2 of the changing role of data analysis with LLMs
Long-form
Dashboards Answer Questions You Already Knew to Ask
Part 1 of 3 on Data Analysis in the age of LLMs
Long-form
Automating Knowledge Graphs because your content needs context
Getting to the basics of knowledge graphs
Long-form
Stop Planning Your AI Strategy. Your Team Already Has One.
Seventy-eight percent of knowledge workers are bringing their own AI tools to work.
Long-form
Agency Branding on a Claude Code Budget. Claude Design in Practice
How much design can you do with AI, when you know nothing about branding?
Long-form
The Rise of Personal Software
Last week I laid out the hidden costs of your SaaS stack — the conformance tax, the integration tax, the simple math showing that adapting to someone else’s software often costs more than building your own.
Long-form
The Economics of a Dying SaaS Market
The average company spends $4,830 per employee per year on SaaS.
Long-form
Surface, Oracle, Ratchet: The Pattern Inside Karpathy’s Autoresearch — Applied to Code
Earlier this month, Karpathy released autoresearch. 630 lines of Python. It ran 700 experiments on a training script over two days, found about 20 real improvements, and the internet lost its mind. 8.6 million views on the announcement.This Substack is reader-supported.
Long-form
SOC2 Ready in a Day: How I Built a Complete Compliance Kit with Claude Code
As part of an ongoing learning, I tackle projects start-to-finish in a day using automation tools.
Long-form
I Live in the Terminal. I Stopped Telling Others to Join Me.
Every morning, I open a terminal and type one command.
Long-form
How to Build a Claw That Won’t Eat You
Claw is about the worst name I can imagine for a product, yet here we are.
Long-form
What My Algorithm Read This Week ( Feb 22, 2026)
I built a system that curates my own newsfeed — not optimized for engagement or ad revenue, just for what’s actually useful to my work.
Long-form
What My Algorithm Read This Week (Feb 15, 2025)
I built a system that curates my own newsfeed — not optimized for engagement or ad revenue, just for what’s actually useful to my work.
Long-form
Trajectory: Teaching Claude Code to Learn Without Touching Model Weights
When you are a developer, everything is a coding problem
Long-form
ReAct Model
An examination of the ReAct pattern for language model reasoning and action, and how it shapes the way we build agent systems.
Field note
Using GPT3 to Build Zero-Shot Recommendation with Whisky Tasting Notes
A practical exploration of zero-shot recommendation systems using GPT-3 embeddings, applied to whisky tasting notes as the test domain.
Field note
Rethinking Search with ChatGPT
How conversational AI changes the mechanics of search, and what that means for the systems we build around information retrieval.