How Slack Keeps a Team of AI Agents From Losing the Plot
Slack’s security investigation service orchestrates a team of large‑language‑model agents—Director, multiple Experts, and a Critic—to analyze alerts without sharing a single monolithic context. The Director records every decision, observation, and hypothesis in a structured journal that feeds into the other agents, while the Critic cross‑checks expert findings for consistency. By partitioning the workflow into three dedicated context channels, Slack avoids token‑limit overflow and the drift that occurs when all agents ingest the full transcript. The design demonstrates disciplined context engineering as a scalable solution for enterprise AI multi‑agent systems.
HubSpot's 37-Minute Lesson in Why HTTP 200 Can Lie
HubSpot’s rollout of a new permissions framework unintentionally omitted role assignments, causing UI workflows for contacts, companies, orders, and projects to disappear for all customers. The access‑control endpoint kept returning HTTP 200 with a restrictive payload, so monitoring systems saw a...
Slack Rebuilt Notifications for Millions of Users
Slack overhauled its notification preferences, merging four fragmented systems into a unified model that separates what to notify from how to deliver. The redesign fixed the misleading “nothing” option that still sent in‑app badges, restoring predictable behavior across desktop and...