Is this AI?
Why the answer tells readers less than they think.
Working Copy
Practical essays on AI, attention, and the future of serious information work.
Live Last read · last published · next check
Latest posts
I Could Not Get AI to Edit One Page
The tools for AI-assisted editing don't exist yet. That’s pushing everyone toward full delegation.
Tools for the Middle of the Work
Against the empty prompt box, part two: five small ideas for steering AI mid-task
Software gardens
What comes after one-size-fits-all apps
The Weird Future of Personal Software
A rough AI-powered feed reader helped me see a different future for software: not finished products, but scaffolds that users and agents grow together.
Against the Empty Prompt Box
AI is good at compressing information into fluent answers. Serious work needs tools that preserve knowledge creation.
Answer Engines Are the Wrong Shape for News
Clients, thick and thin
What happens when the language model lives in the browser?
Testing Gemma 4 for news publishers
Notes on evaluating agentic tools
Unique challenges from complex workflows
Skills Beat Tools
Why teaching agents what you already have outperforms building them something new
Structuring memory
Patterns from the literature
New research: You don’t need the article to know what it’s about
Classifying articles by their URLs
Teaching Agents What We Know
Orchestration and workflows as compound advantages
The year ahead
2026
Private, Free, and Powerful: A Guide to Local AI
Step-by-step walkthroughs to common tasks with small language models
Liquifying Your AI Conversations
A Tool for LLM Data Portability
Building Semantic Search for News: A Technical Teardown of My NLWeb Implementation
A Developer's Guide to Transforming NLWeb from General Purpose to News-Focused
AI Research Assistants: A Comparative Analysis
Testing 7 LLM Tools for Literature Review
New release: Substack API Python package 1.0
Today, I'm excited to announce the 1.0 release of my Substack API Python package. This represents a significant overhaul from the previous 0.1 version, transforming it from a collection of functions into a full-fledged, object-oriented library that makes interacting with Substack's unofficial API more intuitive and powerful.
The LLM prototyping process
Sketching out ideas on wide-open compute