Walk into a great shop and there is always someone there. They greet you. They ask the right questions, listen carefully, and guide you — not just to a shelf, but all the way to the right product. They make the experience feel easy, personal, worth coming back to.

That kind of presence has always been missing online.

For thirty years, e-commerce has worked around its absence. Search bars, filters, product grids, recommendation engines, onboarding wizards — every advance has been an answer to the same workaround question. How do we get the shopper to find the product on their own? The shopper never wanted to navigate. They wanted to be helped.

The newest wave of assistants can finally see the catalog, read the cart, look up an order. That is real progress. But they are still horizontal models pointed at commerce — built to be helpful in general, not trained to sell in particular. They can answer the question. They cannot read the moment.

A great salesperson knows when to offer more, when to suggest a substitute, when to discount to close, when to stay quiet. None of that lives in a system prompt or a function-call schema. It lives in instinct, trained on a particular store's economics — its margins, its customers, its goals. That is the missing layer. And that is the layer we are building.

When AI becomes the interface, everything else changes

The next layer of the internet is not another channel. It is a population of agents, working on both sides of every transaction. The shopper will have one. The store will have one. The marketplace becomes a programmable endpoint between them.

In that world, the merchant's job is no longer to engineer a perfect funnel. It is to give the store a capable representative — one that knows the catalog, the inventory, the customer's history, and the bounds of what it can decide on the merchant's behalf. Reading the catalog is the floor, not the ceiling. The work is judgment — when to push, when to hold, when to discount, when to escalate to a human.

The agent itself becomes the product

A recruiter spends six months training a sourcing agent on healthcare hiring — its candidate signals, its red flags, the kind of conversations that close. That agent is worth renting to every other healthcare recruiter on earth. The instinct, encoded once, can be hired a thousand times.

This is the new shape of software. For decades a product was code that ran in your browser; the work was writing features. The next layer is different. A product is a trained agent — a body of judgment that knows a specific domain and can be hired to apply it. The work moves from writing features to training instinct.

Commerce is the largest version of this story. Every storefront has its own economics, its own customers, its own way of selling. An agent trained on one store is not interchangeable with one trained on another — and that is exactly what makes it valuable. The store does not just buy software anymore. It trains and owns a representative.

And it gets sharper over time. The instinct can be rented; the memory cannot. After two years inside your store, the agent knows your customers — who buys what, which questions close a sale, which substitutions actually work for which goals. Instinct is shared. Memory is yours alone.

An intelligence layer between human intention and economic action.

This is not only technical. It is philosophical

• If agents act for us, how do we encode our values?
• If purchasing becomes delegated, how do we preserve agency?
• If optimization becomes automated, what becomes premium?

We do not believe agents should replace human judgment in commerce. We believe they should give humans more of it — more clarity about what to buy, more confidence that the price is right, more time back. Humans set objectives and boundaries. Agents do the work that used to require browsing.

In short.

Generative shopping systems for a world where agents collaborate and transact — while humans remain in control of objectives and boundaries.

That is the thesis. That is GenCart.