Companion

Companion is an AI assistant built into the BigCommerce control panel, the environment merchants use to operate a store. The product direction was settled; the open question was interaction quality. The engagement answered it the most direct way available: a functioning prototype the team could evaluate and extend, rather than a document describing one.

A prototype in place of a specification

A written specification describes intended behavior and loses fidelity in translation. Here the document was replaced by a running prototype: the assistant live in the admin, handling real queries, exposing the exact interaction engineering would implement. Design decisions were validated in the interface rather than argued on paper.

No dedicated backend

The prototype ran against the platform's existing REST and GraphQL APIs, with the language model orchestrating those calls rather than a new service standing behind it. That kept the build honest about production surface area and inexpensive to iterate, and the interaction could change daily, which is how the product's behavior was found.

The model as the platform's missing search

The merchant admin is deep and dense; core settings are visited rarely and are difficult to relocate. The defining design decision was to use the model as the retrieval layer the platform never had: a natural-language request resolves to the correct screen, record, or figure. Workspaces compress a multi-step administrative task into a single stated intent.