Build a self-updating "What's new" board that turns a busy team announcements channel into a single, always-current digest page. The problem to solve: product and team updates get buried in months of chat history, so nobody can see the current state of things at a glance. Raw posts are noisy and unstructured, and some updates later get removed or renamed, so old announcements go stale. The workflow should fix this by pulling the whole channel, using AI to compile the posts into a clean, dated, categorised digest, and publishing it as a page. Updates that are no longer true should be marked as struck-through or "pulled" so the page always reflects what is currently real. Trigger it on a weekly cron (a full pull and rebuild from scratch each run, not incremental). Build it in these high-level steps: Fetch step — pull the full history of a chat channel and save the raw messages to a shared volume. Use Slack here, or your team's chat tool (Teams, Discord, and similar work too). This should be read-only toward the source. Compile step — have an AI model read all the messages and produce a structured digest: dated sections, each update with a title, a one-line summary, one or more category tags from a fixed set, and a flag for whether the update is still current. Use Anthropic Claude here, or another model provider. Save the result to the shared volume. Give this step a raised timeout since it processes the whole history. Data step — serve the latest compiled digest as JSON over an HTTP route, so other workflows can reuse it. Page step — a webpage that shows the digest newest-first, with filters for topic, date range, and status, plus a view toggle. Struck-through styling for pulled items. Gate the page and data routes behind company SSO so any signed-in employee can view without needing an account. Optionally, add a step that appends each digest into a data warehouse for history and reporting. Use Snowflake here, or your team's warehouse (BigQuery, Redshift, and similar). Design the compile step's output and the data route so the digest is reusable structured data, letting other workflows consume it as an intermediary rather than re-pulling and re-compiling the channel themselves. Tools used: Slack API (conversations.history), Anthropic Claude API, Snowflake SQL API, SSO route authentication