## **Meeting Overview** **Date:** Monday, June 22, 2026 **Attendees:** Aidan, Marcus, Bruno, Emma, Darren, Vidit, Ángela ### **Canvas AI build-state UX** * Generation now runs component by component; editing the canvas mid-build breaks it, so the canvas locks during a build * Build UI: a Stop and a separate Stop-and-undo on the chat, a building indicator, and an optional follow-along scroll * Planning step renders as gray placeholder boxes in the canvas (Lovable-style), with a thumbnail minimap for orientation * Model is a quality/cost/speed dial (Opus best but slow and pricey; mini models fast and cheap); reasoning narration can be toggled off ### **AI Content Review** * The latest-review bug is fixed and merged: a criterion's prior score now persists across revisions * Thresholds per criterion and the content-listing pill are in progress (Eric) ## **Decisions** * Lock the canvas during a build; use Stop and Stop-and-undo as distinct actions * Adopt the glowy-border AI-is-working pattern; consistent with other AI tools over Drupal convention * AI Review keeps prior scores across revisions and labels staleness per row rather than expiring everything * Apply-feedback stays a manual edit-form checklist for v1; Canvas auto-apply waits until Canvas has revisions ## **Questions** * How should a review run on an old revision be labelled per row (revisions-ago plus date), settling the wording Thursday? * Where is report feedback stored so the agent can use it next time, and how are overlapping suggestions outlined consistently? ## **Next Steps** * **Aidan:** Draft a Canvas UI recommendations ticket, then split it; work the revision-label wording Thursday * **UX Team:** Plan / think about Cancel/undo UI that locks * **Bruno:** Content-listing pill then suggestions-on-content; **Eric:** criterion thresholds > Related issue: [Issue #3584833](https://www.drupal.org/node/3584833)