Haiku review: AI-first browser capture, scored 8.38/10

First-draft step titles and body copy came out usable on 11 of 12 test steps, with only one merge fix needed. The summary paragraph read like a human wrote it, not a captions engine. Rewrites happen inline without leaving the doc.

The Chrome extension detects step boundaries automatically as you click through the workflow. Time from starting a capture to a shareable draft was under two minutes on the invoice test. There is no separate render or processing wait.

Reordering, cropping, and annotating are direct-manipulation with a light editor. Brand styling and templated headers are supported but shallower than Scribe or Tango. Bulk operations across long SOPs are a weak spot.

Manual and pattern-based redaction covers the common PII cases. Auto-redaction is present but less aggressive than Scribe's on complex admin screens. SSO and role permissions are available on paid plans.

Public links, gated links, embeds, PDF, and Notion export are all first-class. The exported PDF preserves layout and screenshot crops. Confluence export is embed-based rather than native.

Notion, Slack, Zendesk, and Google Docs export are native. API access is available for the paid tiers. LMS integrations are limited compared to Trainual and ScreenSteps.

Workspaces, roles, and comments cover the basics. Review flows exist but the approval model is lighter than Trainual's. Version history is retained but the diff view is bare.

The free plan is the strongest in the category, covering capture, AI generation, and shareable links. Paid tiers scale sensibly with seat count. There is no upsell wall between capture and export on the free tier.

Time to first published SOP for a first-time user was under five minutes on our test. The UI matches its function: capture, review, share, in that order. No training call needed.