Slack's AI Summaries: Teardown of a Conversation Condenser

AI · 6 min read

Slack introduced an AI-driven summary assistant that produces on-demand thread summaries, decision lists, and action-item drafts. The UI sits as an opt-in banner within threads and in the right-side details rail, with controls to adjust verbosity and a 'confirm before share' toggle for generated action items.

Prompt engineering focuses on roles—participants can request summaries for 'manager', 'engineer', or 'client' framing; each style biases inclusion and tone. This role-based prompt design reduced the number of irrelevant items in summaries but introduced expectations about completeness that required clear disclaimers and easy access to source highlights.

Governance features were baked into UX: admins can disable summaries in marked channels (legal, HR), and every generated summary includes provenance links that jump to the exact message segments used. The design lesson: in-product AI needs explicit controls, provenance, and role-aware prompts to be useful without eroding trust.