# The Example Is the Moral Interface

An example does two jobs at once. It names a mechanism for the builder, and it names a social neighborhood for the user.

That is why Jane Street keeps recurring in my thinking, and why the recurrence needs discipline before it reaches a personal product. Internally, Jane Street is an unusually good reference class for dense feedback loops. Publicly, to many normal users, it is Wall Street with better typography.

Both readings matter.

The useful Jane Street pattern is obvious in Edwin Morris's Claude Code design piece. A designer starts with a problem and proposal, opens the editor, runs the build, gives Claude the written description, proves the functionality in the real codebase, lives with the prototype, pushes it to a development environment, asks users, and submits a feature that already behaves like the intended thing. The prototype becomes a living proposal document. The argument moves out of Figma, prose, and meetings. People can touch it.

That pattern belongs near the center of the Markov Blanket build. A product about crossings should test crossings. A welcome email should be answered by the actual reply loop. A filter rule should be shown as the rule. A voice correction should update the voice object. A user should handle one messy thread and see what the system believes crossed the boundary. The work becomes real when the proposal has behavior.

Jane Street is easier to use this way than other quant firms because it has a public workshop surface. It publishes blog posts, talks, OCaml material, design notes, and culture artifacts that make the inside partially legible. Jump Trading may have equally intense engineering loops; its public technology surface emphasizes hardware, co-location, data centers, AI workloads, and production speed. Renaissance Technologies may be the purest myth of mathematical markets; its public face gives the shape of a quantitative investment company and a scientific campus, then mostly recedes. Jane Street gives outsiders more reusable scenes. It is the quant firm that behaves, from the outside, most like a school.

That explains the obsession. The import has to be translated.

The average first user of an email creature hears finance before microstructure. ETF market making, high-frequency trading, systematic investment management, liquidity provision, risk transfer, lawful local knowledge, and market-manipulation headlines collapse into one social signal. She may hear extraction. She may hear people already rich enough to build brilliant internal tools so they can become richer by moving numbers faster than everyone else. If the product then says, implicitly, "we admire these people," she is allowed to wonder whether the product understands her life.

The blunt version of her prior is tempting to caricature and costly to ignore. Finance people made too much money; the economy feels expensive; smart people sell cognition to capital; if capital allocation is such a public good, why does so much surplus disappear into private compensation while ordinary needs stay ordinary? That sentence is an incomplete theory of markets and a real boundary crossing. A personal AI product that wants access to someone's inbox has to treat it as signal.

The question for the product is therefore separate from the question for market ethics. Price discovery can be productive work. A trade can improve the economy's cognition under real risk, or corrupt it through opacity and rule capture. That boundary still holds. A correct internal theory of price discovery leaves the surface question unsolved. The user is evaluating what the example says about the product's taste.

Examples are trust declarations.

If all the examples come from quant firms, frontier labs, founders, billionaires, and elite workflows, the product teaches the user that it recognizes intensity before care. It may still be benevolent, but benevolence has to fight through the reference class. The user hears that the tool wants to optimize ambitious people. A tool that asks to model a person should make an ordinary person feel like her own life is the starting case.

The same mechanism can be translated while preserving its force. Jane Street's "prototype as living proposal" becomes: send one stressful thread and receive a concrete reply you can correct. The trading desk's rapid feedback becomes: try the rule on one mailbox slice before connecting everything. Code review becomes: the model's answer is a proposal, and the user's correction is the authority. Risk management becomes: human review stays on outward messages until the trust boundary has earned motion. The underlying loop survives. The social neighborhood changes.

This is the design rule: keep two ledgers for examples.

The internal ledger is allowed to be alien, elite, technical, and morally complicated. It can contain Jane Street for executable proposals, Jump for speed close to hardware, Renaissance for mathematical investing, SpaceX for physical cadence, Stripe for developer trust, Berkshire for allocation discipline, and whatever other machine teaches the build team how to think. The internal ledger optimizes for mechanism density.

The surface ledger has a different job. It should begin where the user's boundary already hurts: a school email, a medical bill, a work thread that needs a careful answer, a volunteer schedule, a parent-teacher exchange, a landlord note, a friendship she has been avoiding, a small-business customer reply, a forwarded article that irritated her for reasons she cannot name yet. The surface ledger optimizes for moral proximity.

A product that can hold both ledgers has better taste than a product that chooses only one. Internal examples without surface translation become alien. Surface examples without internal machinery become nice copy around a weak loop. The point is to learn from the densest machines and speak first from the user's life.

This also protects collaboration. Morris names the downside of executable prototypes: a reviewer can receive a fully baked feature and feel reduced to code review, when the prototype was supposed to invite design feedback. The same failure appears in personal AI. If the model returns a polished account of the user's life, the user may feel diagnosed from above. The answer has to arrive as a living proposal: here is what I think crossed your boundary; here is the rule I inferred; here is the draft I would send; correct me.

That sentence is the product posture.

Jane Street belongs in the workshop because it shows how proposals become behavior. It belongs in provenance because it explains the build loop. It belongs in public-facing examples only when the user is already asking about that world. For everyone else, the first example should be the thing she actually opened the product to solve.

The example is the first model of the user that the product reveals. Before the system reads her mail, before it drafts in her voice, before it claims to understand her why, it chooses a story about who this is for. That story is already an interface. It is already a moral claim.

Learn from Jane Street. Greet the user at her own inbox.

## Source Notes

- Edwin Morris, Jane Street Blog, ["I design with Claude more than Figma now"](https://blog.janestreet.com/i-design-with-claude-code-more-than-figma-now-index/), 2026-02-05.
- Jane Street, ["What We Do: Overview"](https://www.janestreet.com/what-we-do/overview/), for the public claims about liquidity provision, trading on 200+ venues, OCaml, and porous trading/research/technology teams.
- Jump Trading, ["Technology"](https://www.jumptrading.com/technology), for the public claims about hardware, co-location, large-scale AI/ML workloads, and fast production feedback.
- Renaissance Technologies, [careers brochure](https://www.rentec.com/pdf/careers_brochure.pdf), for the public description of the firm as quantitative investment management founded by James Simons and using mathematical and statistical methods.