Product work across strategy, design, and build.

A separate product, with a migration path.

I explored adding the agent to XOIP and worked through the design. It did not hold up. The existing product relies on simplicity through constraint; an AI agent needs room for more open-ended behavior. A demo made the difference clear to stakeholders.

The setup should be a conversation.

Asking SME owners to set up an AI agent through forms and toggles is the wrong model. A conversation should be the primary interface. Settings can show or change details when needed.

Structured views make the result easy to check.

Chat drives configuration. SME owners still need to check what they have set up. Routing rules and the directory remain as read-back views: the conversation creates the data, and the panels make it easy to inspect.

The agent speaks for the brand.

A law firm should sound different from a surf shop. The persona picker lets owners choose and audition a named voice that suits their business, so the agent feels closer to a colleague than a generic settings screen.

Every call is a test.

Each call is an implicit test. An automated eval layer performs deterministic checks and has AI assess quality on every interaction. It turns findings into cards with concrete next steps. Next, those cards will go to coding agents for automatic remediation.