AI Writer & Message Intelligence

Two product designers. I owned the writer interaction model, verification patterns, and the summary's desktop constraint — persistent panel to popover. We shared research and reviews. I recommended parking the local assistant when overlapping initiatives and limited engineering made it the wrong bet.

Existing research on AI in RingCentral Messaging was consistent: people questioned accuracy, wanted help with tedious work, and could not see a clear benefit. A separate AI destination would have asked them to trust a black box.

I designed verification into the task: suggestions replace a draft only when accepted, summaries can be regenerated or copied, and an AI disclaimer made the source of the output explicit. Control stayed with the person sending the message.

Writer sits in the message input. Users pick a refinement, see a result they can keep editing, and accept it back into the draft. Summary starts from the unread badge — the moment someone is already trying to catch up — not from a new nav item.

Competitor tools (Type AI, Grammarly, Slack, Copilot) confirmed the pattern: entry next to the work, not a separate product. We did not need a gallery of their screens to make the call.

The desktop summary was designed as a persistent side panel so people could read catch-up and keep the conversation in view. Engineering could not support that surface in time.

I moved it to a popover rather than delay the feature or fake a panel the stack could not ship. On mobile, the opposite pressure: a cramped overlay became a full takeover so the content was readable. The interaction principle stayed — stay in the conversation — the chrome changed to match what could actually launch.

The assistant was a conversation-aware chat from the header: suggested prompts, cited answers, tied to the same summary model. It was the most destination-like of the three surfaces.

The local assistant was no longer the right product bet. Engineering capacity was shared with Phone, global AI chat was already underway, and Writer plus Summary could deliver value without introducing another assistant surface. I recommended stopping the assistant and preserving the trust and evaluation patterns that had already proven useful. Knowing when not to build is part of the job. We still took it through enough iteration to know what we were putting down.

A product direction stopped, but the design and research framework created lasting value. After handoff I stayed through UI verification: workarounds when implementation drifted, then an internal beta. The beta produced concrete changes — summary entry moved to raise discovery, branding adjusted, disclaimer added — not a vague “positive” result.

Writer and summary were prepared for general availability, then the initiative was deprioritized in a broader platform shift. The durable artifact is the evaluation frame: behavior, trust, task efficiency, discoverability, qualitative signal, and business impact. Adoption tells you people tried it. Those six dimensions tell you whether to keep it.