Economists classify goods by when the buyer can judge their quality. In 1970 Nelson distinguished search goods from experience goods. The quality of a search good can be judged before purchase by inspection and comparison, as with computers and clothing; the quality of an experience good is known only after use, as with restaurants and haircuts. In 1973 Darby and Karni added a third category, credence goods: even after use, the buyer finds it hard to judge their quality, as with car repairs and medical diagnoses.
Financial products are credence goods, and among the hardest of them to judge. The buyer pays with money today and receives in return a promise about the future: after how many years, in which circumstances, from whom, and how much. The outcome often arrives many years later; even when it does, it is hard to tell whether it came from the product itself or from luck.
The change brought by large language models is more than an extra chat box in the interface. Once AI can understand intent, call tools and carry out tasks over time, a financial product can for the first time act on the user's behalf: read financial reports for the user, follow the market for the user, rebalance for the user. The product's role changes from tool to agent.
From tool to agent
The link between user and market changes from following orders to deciding for itself.
Designing a financial product requires understanding at the same time how markets form prices, how users make decisions, and how promises are kept credibly. In the past these three belonged to three kinds of people: investment research, product, and compliance. While product forms were stable, each could handle its own segment and hand over at the boundary. Once an agent carries the user through the whole process, every step needs all three disciplines at once.
From handovers to all present at once
The five stages an investment passes through, and which disciplines each stage needs.
Those who know only AI tend to overestimate what an agent can do. Those who know only product tend to treat finance as an ordinary consumer setting. Those who know only finance tend to underestimate how much user behaviour changes once the mode of interaction changes. This book discusses the three within a single framework.
On financial principles, the book follows the line that runs from Bachelier and Fama to multi-factor scientific investing:
Corollary: where AI adds value in financial products
not in predicting prices or picking winners
This stance is chosen not because it explains everything, but because it is the framework with the strongest evidence today and the most honest one for ordinary investors. It also directly determines where the value of AI in financial products lies.
How markets absorb information, who provides liquidity and who bears the cost of trading are questions for the study of market microstructure. Here the book draws mainly on O'Hara's Market Microstructure Theory and Harris's Trading and Exchanges: Market Microstructure for Practitioners.
The book is written for two kinds of reader. One is the practitioner newly entering the financial products industry, who needs a complete map. The other is the experienced practitioner who is already deep in one discipline and still has gaps in another.
Choose a reading order by background
Pick a background: solid circles are disciplines you already have, dashed circles are ones to fill in.
Plain narrative, thorough citation, rigorous logic. Each chapter starts from an everyday, observable phenomenon and assumes no professional background. Yet every claim is either supported by published research, with author, year and source given, or is reasoning that can be checked. Concepts use the standard terms of their discipline and are defined on first use.
Visualization first, leaving the conventional book behind. Most of the logic of finance consists of dynamic mechanisms: change one condition and the conclusion changes with it. How volatility erodes long-term returns, why a market maker widens the spread when informed traders are numerous, how a market shrinks when buyers cannot tell good from bad: these processes are abstract in long passages of prose and too difficult in formulas. The book hands the mechanisms to figures, animations and simulations the reader can operate.
Four ways of explaining
Content of different kinds goes to different forms of visualization.
Structure → diagram
Who is connected to whom, and where each one sits
Process → animation
What happens first, what follows, and how it is passed on
Parameters → simulation you can operate
Change one condition and the conclusion changes with it
Randomness → distribution
Show every possibility, not just the average
Every substantial section has a figure. If a section cannot be drawn, it is fair to doubt whether it can really be understood: either remove it, or find a way to draw it clearly. The text poses questions, connects the argument and states conclusions; the figures present mechanisms. Animation is used only to show processes, never as decoration; the text reads through without the animations.
Economic analysis as the main thread, with as little mathematics as possible. Every question starts from the participants: who takes part, what their motives and constraints are, and what outcome results. Where a formula would be needed, a visualization takes the place of the derivation.
How this book was made. The author set the book's purpose, its structure and the direction of each chapter's argument. Claude wrote the first draft of the text and the interactive figures; the author reviewed, rejected and revised them chapter by chapter, and some chapters draw on articles the author has published. Claude checked every citation and number in the book against public sources, and the author reviewed the checks; claims whose source could not be found are not used as evidence. A book about how much AI can do on people's behalf is itself one such delegation.
If this book saves a newcomer a few years of detours, or gives an experienced practitioner the perspective of another discipline, it will have served its purpose.