Slack · Supporting pattern
Multi-agent context system
Slack's engineering team runs an agentic service that investigates security alerts. A Director agent orchestrates rounds, Expert agents gather evidence, and a Critic scores each finding for credibility. Instead of message history, three written channels carry the state between rounds.
This entry describes supporting infrastructure that other work builds on. The catalog classifies it as a supporting pattern. The record also reports a workflow.
- Approach type
- Supporting pattern
- Work
- Research
- Human involvement
- Human in loop
- Deployment stage
- Research
- Evidence strength
- Mixed
- Entry reviewed
How it works
The workflow the sources report for this implementation.
Structured working memory: findings, decisions, questions, hypotheses
A truth filter with credibility scores over submitted findings
A chronological, deduped, conflict-resolved synthesis retained across steps
Where people stay involved
- long-running investigation → synthesized reportUnknown · Level unknown
Unclassified for long-running investigation → synthesized report; human attention boundary: unknown.
- Observation date
- 2026
Implementation details
Coordinator/dispatcher: a central coordinator dispatches to expert agents and to critic agents
Not specified
Expert agents produce reports; critic agents evaluate them using evidence-inspection tools
Three channels; Director's Journal (working memory), Critic's Review (credibility-weighted findings), Critic's Timeline (deduped chronological synthesis)
Reported results and limitations
The catalog records what the sources report, with the scope and the denominator of every figure. A qualification below limits the figure it sits under.
Reported metrics
Context management for security investigations spanning hundreds of inference requests and megabytes of output
The source does not report the denominator of this figure.
- Reported by
- Slack
- Scope
- Complex security investigations requiring tailored multi-agent context; qualitative workload scale
Handles multi-agent runs spanning hundreds of requests and megabytes of output
The source does not report the denominator of this figure.
- Reported by
- Slack
- Scope
- Complex security investigations requiring tailored multi-agent context; qualitative workload scale
Lessons and interpretation
Don't pass all information at every step; build structured summaries agents can reliably build on
Separate expert agents (produce) from critic agents (evaluate); corroborated findings are prioritized
Context management becomes its own subsystem once runs get long
Sources and research details
Citations link to the original publisher. Each source also keeps a preserved copy in the repository, so a changed or removed page stays checkable.
- Managing context in long-running agentic applicationshttps://slack.engineering/managing-context-in-long-run-agentic-applications/
- How Slack manages context in long-running multi-agent systemshttps://www.infoq.com/news/2026/04/slack-agent-context-management/
Research details for every claim on this page
- Summary
- Statement type
- Fact
- Provenance
- Reported
- Confidence
- High
- Confidence reason
- A linked first-party source states the claim.
- Headline claim
- Statement type
- Metric
- Provenance
- Reported
- Confidence
- High
- Confidence reason
- A linked first-party source states the claim.
- Reported by
- Slack
- Scope
- Complex security investigations requiring tailored multi-agent context; qualitative workload scale
- Denominator
- Not reported
- Method
- Not reported
- Observation date
- Not reported
- SupportsManaging context in long-running agentic applicationsPreserved content.md, lines 34–38
- SupportsHow Slack manages context in long-running multi-agent systemsPreserved content.md, lines 22
- Harness
- Statement type
- Fact
- Provenance
- Reported
- Confidence
- High
- Confidence reason
- A linked first-party source states the claim.
- Model
- Statement type
- Fact
- Provenance
- Reported
- Confidence
- High
- Confidence reason
- A linked first-party source states the claim.
- Tool access
- Statement type
- Fact
- Provenance
- Reported
- Confidence
- High
- Confidence reason
- A linked first-party source states the claim.
- Context management
- Statement type
- Fact
- Provenance
- Reported
- Confidence
- High
- Confidence reason
- A linked first-party source states the claim.
- Supporting component
- Statement type
- Fact
- Provenance
- Reported
- Confidence
- High
- Confidence reason
- A linked first-party source states the claim.
- Supporting component
- Statement type
- Fact
- Provenance
- Reported
- Confidence
- High
- Confidence reason
- A linked first-party source states the claim.
- Supporting component
- Statement type
- Fact
- Provenance
- Reported
- Confidence
- High
- Confidence reason
- A linked first-party source states the claim.
- Key observation
- Statement type
- Metric
- Provenance
- Reported
- Confidence
- High
- Confidence reason
- A linked first-party source states the claim.
- Reported by
- Slack
- Scope
- Complex security investigations requiring tailored multi-agent context; qualitative workload scale
- Denominator
- Not reported
- Method
- Not reported
- Observation date
- Not reported
- SupportsManaging context in long-running agentic applicationsPreserved content.md, lines 34–38
- SupportsHow Slack manages context in long-running multi-agent systemsPreserved content.md, lines 22
- Lesson
- Statement type
- Inference
- Provenance
- Catalog judgment
- Confidence
- Medium
- Confidence reason
- The catalog derives this observation from the linked sources.
- Lesson
- Statement type
- Inference
- Provenance
- Catalog judgment
- Confidence
- Medium
- Confidence reason
- The catalog derives this observation from the linked sources.
- Lesson
- Statement type
- Inference
- Provenance
- Catalog judgment
- Confidence
- Medium
- Confidence reason
- The catalog derives this observation from the linked sources.
- Operating model assessment
- Statement type
- Inference
- Provenance
- Catalog judgment
- Confidence
- Unverified
- Confidence reason
- The research pattern documents agent coordination and criticism but not the normal human attention boundary.
- Observation date
- 2026