The latest research direction within the Software Agents group of the MIT Media Lab explores a scenario called "Impulse" in which e-commerce meets 'brick-and-mortar' commerce through a system of buying and selling agents representing individual buyers and sellers that engage in multi- parameter negotiation and run on wireless mobile devices.
The InShop project is an initial realization of this Impulse scenario, developed to enhance the grocery store experience. InShop consists of several applications running on different types of wireless devices. In the InShop scenario, while a consumer uses a hand-held device to scan items in a supermarket, software agents work in the background to assist the user with shopping hints and recommendations.
The InShop software agents work proactively on behalf of the user's interests by using information in the user's pre-existing shopping history and by observing which items the user scans and places in his/her cart. In this manner, the agents supply just-in-time information in the form of item, coupon and recipe recommendations.
The InShop applications are built upon a distributed agent architecture called Hive. Hive allows agents to reside and communicate across wireless devices, database servers, and desktop machines. This distributed infrastructure enables flexibility in the final devices used in the store.
The three main InShop recommendation agents are:

Coupon Finder - When you enter the store, based on your shopping list and items that you frequently buy, you can collect electronic coupons which will be stored on your hand held device.

Item Recommender - As you move through the store picking up things on your shopping list or selecting other items, your agent will recommend new and related items you may want to purchase.

Recipe Recommender - When you scan an item in the store, your agent supplies a list of recommended recipes based on what kind of user you are and what items you've just placed in your cart.