Managing context in long-run agentic applications
Slack describes three complementary context channels—a Director's Journal, a Critic's Review with credibility scoring, and a Critic's Timeline—that let teams of agents stay coherent across long-running security investigations without passing full message history between rounds.
More from Slack Engineering
Streamlining Security Investigations with Agents
Slack's Security Engineering team deployed a multi-persona agent system — Director, Expert, and Critic agents working together — that performed over 7,500 security investigations and 500,000 tool calls in its first quarter, with structured outputs reducing hallucinations.
Empowering Engineers with AI
Slack describes deploying AI-powered developer tools that save 10,000+ hours of engineering time yearly. Their internal AI assistant has been generally available for five months with 4,000+ monthly chat-turns and 70%+ positive satisfaction across the engineering organization.