Drafted from real merges. Edited by humans.

AI answers, streamed alongside a visible plan

Help-center search now generates inline AI answers when you ask a question. Streams a 'Searching → Reading → Answering' plan before the tokens — the wait-tolerance trick from Perplexity — with inline citation chips that link back to the source articles. Article result rows also pick up a 96×54 walkthrough thumbnail, the bespoke move competitors without recordings can't match.

• Inline AI answer above results when the query parses as a question
• Plan stream + token stream via Claude Haiku and Server-Sent Events
• Numbered [n] citation chips matched to a top source strip
• Walkthrough thumbnails on every help-center result row

Hero search on the help center

A big centered search bar at the top of /help-center with a Raycast-style live two-pane preview — results on the left, focused-result preview on the right, keyboard-driven scrub. Results are cross-surface: help center plus dev docs plus changelog plus blog in one query, so visitors get the complete picture in one shot.

• Big hero search input with popular-query chips
• Two-pane overlay: results left, live preview right
• Keyboard-driven (arrow keys, Enter, Esc)
• Cross-surface mosaic — one query spans all four content surfaces

Why this blog is written by the same pipeline that writes yours

We could hand-write a blog. We don't. The posts on this page come from the same merge-to-draft pipeline that ships customer changelogs and walkthroughs — the same drift detection, the same review queue, the same operator-edits-before-publish loop. If the product can't write the product's own blog, it shouldn't be writing yours either. This is the canary, and the canary is alive.