Start Here

These essays explain AI from the mechanism outward. Some stay inside the model: what it reads, how training shapes it, why it is confidently wrong. Some follow the model into the software around it, into a team's work, a company's budget, a market's structure, a person's job. Each one starts from a question you can recognize and goes as deep as the question needs.

You don't need to read them in order. Every piece explains what it depends on in a sentence or two and links to the essay that develops it. Terms that tend to be assumed, starting with model, are linked to short definitions the first time they matter.

If you read only three

• Same Model, Different Results: why the software around a model moves results as much as the model, and which parts of an AI system to delete with each release.
• Where the Saved Time Goes: why making one step ten times faster rarely makes the job much faster, and what does.
• The Model Never Sees Your Words: what a model actually receives, measured across 30 languages, and what follows for cost, context and quality.

Routes by what you need

I use AI and want to get better at it. Start with Learn to Drive Before You Study the Engine, then The Model Never Sees Your Words for the one piece of mechanism that explains the most odd behavior, Same Model, Different Results for why the same model differs between tools, and Confidently Wrong for when to check.

I build AI products and systems. Same Model, Different Results sets up the vocabulary (harness, scaffolding, walls). Then How a Coding Agent Actually Works, Build an Evaluation From Your Own Work, When the Tool Call Fails Halfway, Write for the Model That Doesn't Exist Yet and What Fits Versus What Gets Used. A Product That Improves With the Model turns the same ideas into product design: which products gain from each model release and which get absorbed.

I lead a team or run an AI program. Start with Designing an AI Program That Ships. From a Good Demo to a Working Department and Data Access Is the Real Integration Project cover what decides whether a deployment works. Why the Process Exists and Moving Without Breaking cover changing processes without losing what they did or stopping the operation. From Individual Gains to Team Capability and What a Deployment Engineer Should Own, and Leave Behind are about people and roles, and Measuring AI Impact Without Fooling Yourself about knowing whether any of it worked.

I want to understand how models work. What Does a Model Actually Learn?, What Training Actually Changes, The Model Never Sees Your Words, What Thinking Buys and Why Progress Arrives Unevenly.

I decide about money, markets and strategy. Where the AI Money Goes, Who Should Build It, Where the Moat Moves, The Price of Know-how, Why Labs Build Products and Who Receives the Gain?.

I want to know what this means for people and jobs. A Job Is a Bundle of Tasks, Customer Support Went First, What Changes When Code Gets Cheap and Learn to Drive Before You Study the Engine.

How to read the evidence

Every essay ends with its sources, and pieces that measure or compute something include a Method section explaining how, so you can check or rerun it. Figures carry a small label: measured (data from a stated source), computed (derived from stated assumptions or formulas) or conceptual illustration (a diagram of a relationship, with no data behind it). Volatile facts such as model names, prices and benchmark scores carry the date they were checked. Dated notes, such as When Test Agents Touch the Real World, explain a recent event through a mechanism and say when they were written.

Where an essay states a hypothesis, it says so and names what would change the conclusion. Judgments about priorities are argued, not presented as findings.

What is still being written

The collection is a work in progress. Pieces linked above but not yet available are marked as in preparation. The map behind them covers models and reasoning, programming and software architecture, context and memory, agents and standards, evaluation, infrastructure economics, product and distribution, enterprise adoption and transformation programs, team organization, value and impact, buying and partnering, labs and startups, durable advantage and acquisitions, industries, people and work, regulation and second-order effects.