Weave AI

The Problem

New customers were struggling to complete required setup tasks during onboarding. This slowed time to value and increased reliance on onboarding specialists and account executives.

At the same time, AI experiments were emerging in isolated product areas. These efforts were scoped to individual features and lacked a unifying platform vision. Without convergence, this risked duplicated infrastructure, inconsistent interaction models, and long-term technical debt.

I saw two related problems:

1. Onboarding friction

2. Fragmented AI strategy

The Insight

Early in exploration, I assumed AI work in Messaging would evolve into a platform-level assistant. Feedback revealed that this was not the long-term plan.

That changed the direction of the project.

Instead of designing an onboarding chatbot, I reframed the opportunity:

If AI is going to exist across the product it needs to be designed as infrastructure, not a collection of features. We have to establish a North Star to guide early designs.

Onboarding became the entry point, not the end state.

Defining the Platform Vision

I created a long-term AI vision that included:

• High-value use cases across onboarding, messaging, payments, and settings

• A unified interaction model

• A clean evolution path from onboarding assistant to global AI worker

• Design principles for embedding AI naturally within workflows

• Shared context and memory layer across AI touchpoints

This vision clarified:

• What we were building now

• What we were enabling later

• What infrastructure would be required

Driving Alignment

To move beyond concept, I led conversations across:

• Product leadership

• Engineering leadership

• ML team

• Integrations team

These discussions focused on:

• Data access requirements

• Model ownership

• Integration architecture

• Escalation patterns to human support

• State persistence

The goal was alignment before execution. This shifted the conversation from shipping an onboarding feature to sequencing platform capabilities intentionally.

Designing the Onboarding Assistant (MVP)

With alignment in place, I designed the AI Onboarding Assistant as Phase 1 of the broader system. The assistant was intentionally scoped to validate both the interaction model and the underlying AI infrastructure required for broader expansion.

Key decisions:

• Contextual entry points within onboarding tasks

• A persistent but non-intrusive assistant

• Clear escalation to human support

• Reusable interaction patterns

• State-based visual system

I named and branded the assistant, designing an icon that could communicate:

• Idle

• Listening

• Processing

• Responding

• Error states

I also created motion studies to define how the assistant would behave.

The visual system was intentionally extensible so it could scale into a global assistant without redesign.