Turning a blank AI chat window into a configured workspace, thirteen plugins deep

Knowledge workers open an AI assistant to a blank screen. It knows nothing about them, their clients, their voice or their files — so they try it, get generic output, and conclude AI doesn't work for their job. The gap isn't model capability. It's that nobody configured it.

We built a commercial product line of 13 plugins: one foundation plugin that configures the workspace, and 12 profession-specific packs covering accountants, real-estate agents, insurance brokers, financial advisers, management consultants, marketing agencies, nonprofits, ecommerce operators, product managers, sales directors, business coaches and creative freelancers.

The foundation plugin runs a 13-question guided interview built as a conversational state machine — one question at a time, with a documented response framework, a protocol for thin answers, another for answers given out of order, and a resume path for sessions interrupted mid-way. From that it classifies the user into one of four working archetypes and generates their entire workspace: a personalised instruction file, a folder architecture, a persistent memory system, a connectors manifest, and a first task designed to produce a visible win immediately.

Each persona pack then layers domain expertise on top — 20 to 24 named capabilities across five categories, so a configured workspace knows what work that profession actually does.

The engineering discipline is what makes it a product rather than a prototype: strict version synchronisation across three files per plugin, a licence shipped in every pack, and per-user package baking with a recurrence guard that prevents intellectual property leaking between customers.