The Super-User Paradox
The AI productivity paradox: super-users are getting faster, but most companies still do not turn AI gains into revenue, cost, or operating performance.
Context engineering for professionals
The AI-native professional
The professionals pulling away are not using AI more. They are teaching it how they think, decide, and communicate, then compounding that advantage across real work.
Pick a guide that matches a problem you have today. Subscribe if you want the next one in your inbox.
Start here
Begin with the lead guide, then explore the queue.
How to Read an AI Benchmark
Why the labs keep retiring their own headline tests, and how working professionals should actually read frontier-AI launch numbers in 2026.
Shipping a Multiplayer Game in 4 Days: A Spec-First Build Log
A non-engineer shipped a production multiplayer game in four days using spec-first thinking and Claude Code. Here's the full build log.
Latest reading
See all articles
I Built a Paul Graham AI Profile
What happens when you load 1,875 tokens of structured context into Claude? A first-person experiment in context engineering with a Paul Graham AI profile.
The Multiplier, Not the Equaliser
Ben Horowitz says AI equalises opportunity. The economics of data, power, and capital suggest a harsher truth: AI multiplies existing advantage.
Why I Manage My AI's Memory
ChatGPT's memory knows your name. It doesn't know how you think. The Two-Layer Context Model explains why, and what to do about it.
I Spent Days Trying to Master AI Image Generation. Here's What Actually Works.
A five-step AI image generation prompt engineering framework built through real editorial illustration work. ChatGPT vs Gemini, tested and compared.
More from Learned Context
Essays land each week. Subscribe to get them in your inbox, or read the running index.
01
Weekly dispatch
One sharp essay and curated links each week, written for operators who take their AI seriously. Free.
02
About
Why Learned Context exists, who it is for, and the working method behind the publication.
Foundations
Browse topicI Built a Paul Graham AI Profile
What happens when you load 1,875 tokens of structured context into Claude? A first-person experiment in context engineering with a Paul Graham AI profile.
Teaching an AI to Write Like You (Not Like Every Other AI)
Voice calibration is harder than it looks. Rules alone hit an 85% ceiling. The three-layer architecture that gets closer.
The Architecture of Teaching AI Your Job
AI that understands your job needs three layers of context: cognitive, professional, and operational.
AI Memory vs. AI Context: What Professionals Need to Know
AI memory saves facts about you. AI context captures how you think. The difference determines output quality.
Playbooks
Browse topic
The Working Professional's Guide to Claude
From Chat to Code: you don't need to be technical to use the most capable AI tool on the market. How to start with Chat, graduate to Cowork, and decide whether Code is worth the leap.
The Quiet Frontend Gap: Claude Code vs Codex for Web Development
Claude Code dominates backend and architecture. Codex wins routine frontend. A practitioner's analysis of where each excels.
Your AI Knows Your Name. It Doesn't Know How You Think.
AI memory features remember your preferences but not your professional reasoning. Why that gap matters.
How to Stop AI Hallucinating at Work
AI hallucination is a context problem, not just a model problem. How structured context constrains confident nonsense.