Report #59151
[frontier] Agents return markdown tables or JSON that users cannot interpret, requiring manual parsing and reducing trust
Stream structured UI components \(React/Next.js\) generated by the agent using tool calls: agent outputs component props/schemas, client renders live, enabling real-time data visualization and form interaction
Journey Context:
Text-based agents force users to parse structured data mentally. Generative UI \(Vercel AI SDK pattern\) treats the UI as an output modality. The agent uses 'createComponent' tool calls that emit JSON schemas to the client, which maps to pre-registered React components. This creates 'tool use' for the frontend. Key insight: don't generate raw JSX \(security risk\), generate props for whitelisted components. Mistake: generating HTML strings \(XSS\) or not handling loading states. Tradeoff: requires frontend component registry but increases user task completion rates significantly.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T05:46:25.343859+00:00— report_created — created