Skip to main content

AI Strand – The Collaborative Intelligence Layer

AI is not a feature — it’s the intelligence layer of your company. The AI Strand defines how your company:
  • understands and summarizes information,
  • augments search and discovery,
  • automates workflows and decisions,
  • powers agents that act on behalf of users,
  • and does all of this safely, explainably, and under governance.
This is your AI OS: the blueprint for how intelligence is applied everywhere in the product.

🧪 Workshop Meta – How to Design the AI Strand

Framework version: ai-strand-v1.0 Templates this strand covers
  • AI Purpose
  • AI Use Cases
  • AI Surfaces
  • Model Architecture
  • Data Inputs
  • AI Reasoning Abilities
  • Integration with Product Strands
  • Automation System
  • Agents & Bots
  • Privacy & Security
  • Governance & Guardrails
  • AI Performance Metrics
  • Human + AI Collaboration
Who should be in the room
  • AI/ML
  • Data
  • Engineering
  • Product
  • Security & Compliance
  • Support / CX
Facilitation notes
  • Start by reverse-engineering actual AI features:
    • search,
    • summaries,
    • workflows,
    • automation,
    • agent actions.
  • This becomes the AI OS — how intelligence is designed, deployed, and governed across the company.

🎯 Purpose & Role – Why AI Exists

Guiding question
Why does AI exist inside Slack?
Core answer AI amplifies human work by:
  • summarizing information,
  • accelerating search,
  • automating repetitive tasks,
  • understanding intent,
  • enabling agents to interact with business systems on behalf of users.
AI transforms Slack from a communication tool into a collaborative intelligence layer. Objectives
  • Reduce cognitive load in complex, high-velocity communication.
  • Surface the right information at the right time.
  • Automate workflows that previously required manual coordination.
  • Enable teams to work asynchronously without losing context.
  • Let users interact with enterprise systems via natural language.

🧠 AI Use Cases – Where Intelligence Shows Up

Guiding question
What jobs does AI actually do inside the product?

1. Information Understanding

Capabilities
  • Channel summaries (daily or on-demand).
  • Thread summaries.
  • Meeting / huddle summaries.
  • Canvas summarization.
  • Long-file summarization (PDFs, docs, etc.).

2. Search Intelligence

Capabilities
  • Semantic search powered by embeddings.
  • Reranking search results based on relevance.
  • Federated search across files, messages, and integrated systems.
  • Query rewriting and query understanding.

3. Workflow Intelligence

Capabilities
  • Suggest workflow automations.
  • Auto-fill workflow steps.
  • Extract structured data from conversations.
  • Trigger actions based on patterns in channels.

4. Agent Interactions

Capabilities
  • AI agents that execute tasks:
    • create tickets,
    • update CRM,
    • schedule meetings.
  • Multi-step reasoning to interact with APIs.
  • Slack-native bot personas and skills.
  • Agents that collaborate with each other in shared channels.

5. User Assistance

Capabilities
  • Draft message generation.
  • Rewrite for clarity, tone, and conciseness.
  • Explain complex topics.
  • Translate messages.
  • Generate templates (announcements, standups, updates).

6. Security & Compliance

Capabilities
  • Sensitive data detection.
  • Risky user behavior alerts.
  • Automated compliance checks.

🧩 AI Surfaces – Where Users Touch AI

Guiding question
Where does AI appear inside Slack?

Primary Surfaces

  • Search bar – AI semantic search + summaries.
  • Message composer – drafts, rewrites, tone changes.
  • Message actions – “summarize thread”, “summarize channel”.
  • Home tab – AI insights and workflows.
  • Workflow Builder – AI step suggestions.
  • Canvas – AI tooling for documentation and synthesis.
  • Files – AI summaries of PDFs, docs, images.
  • AI sidebar – concierge, queries, and agent orchestration.

🏗 Model Architecture – How AI Is Built Under the Hood

Guiding question
How is Slack AI architected?

Components

  • Slack AI Core
    • Controls routing, memory, context packaging, and guardrails.
  • LLM Layer
    • Models:
      • Salesforce Einstein models.
      • Partner LLMs (OpenAI, Anthropic, Cohere, etc.).
      • Fine-tuned domain-specific models.
    • Notes:
      • Model is selected by task type, data sensitivity, cost/latency.
  • Embedding Layer
    • Role:
      • Generates semantic vectors for search and summarization.
    • Stores:
      • Vector DB embedded inside Slack search architecture.
  • Context Management
    • Role:
      • Retrieval, compression, and chunking of messages, threads, and files.
  • AI Execution Engine
    • Role:
      • Runs agents, multi-step workflows, and action sequences.

📥 Data Inputs – What AI Sees (and What It Doesn’t)

Guiding question
What data does AI use and how is it controlled?

Data Sources

  • Messages.
  • Threads.
  • Channels.
  • Files.
  • Canvas documents.
  • Lists / projects.
  • Workflow logs.
  • Telemetric signals.
  • User preferences.

Privacy Controls

  • AI only accesses data users already have permission to view.
  • Workspace-level admin controls for AI features.
  • Enterprise Key Management (EKM) for encryption.
  • Model and data routing based on compliance region.

Excluded Data

  • Admin-only messages.
  • Private security logs.
  • Restricted channels unless the user is a member.

🧩 Reasoning Abilities – What AI Actually “Thinks” About

Guiding question
What cognitive tasks does Slack AI perform?
Abilities
  • Summarization.
  • Classification.
  • Semantic retrieval.
  • Intent detection.
  • Context distillation.
  • Step sequencing for workflows.
  • Natural language → structured logic mapping.
  • Query rewriting.

🌐 Integration with Other Strands – AI Everywhere, Not One Team

Product Strand

  • AI features embedded into:
    • channels,
    • search,
    • workflows,
    • canvas.

UX Strand

  • AI onboarding flows.
  • AI explainability surfaces.

UI Strand

  • AI buttons.
  • AI contextual menus.
  • AI sidebar.

Data Strand

  • Uses semantic vectors stored in search index.
  • Uses full message + file corpus (with access rules).

Support Strand

  • AI agent for first-tier support.
  • AI-suggested drafts for human agents.

Operations Strand

  • AI-based incident routing.
  • AI logs for anomaly detection.

Sales Strand

  • AI-generated ROI stories.
  • AI surfaces usage insights for sales reps.

🤖 Automation System – How AI Drives Automation

Guiding question
How does AI power automation inside Slack?

Architecture

  • Workflow Builder + AI (no-code automation creation).
  • Bots triggered by message patterns.
  • Agent-based automations.
  • Scheduled digests and summaries.

Automation Types

  • Notifications based on events.
  • Approvals and routing.
  • Incident alerts.
  • Data extraction from conversations.
  • Running scripts or API calls.

Governance

  • Admin rules for workflows.
  • AI cannot execute destructive actions without confirmation.
  • Rate limiting on automations.

🧿 Agents & Bots – AI That Acts on Your Behalf

Guiding question
How do AI agents work inside Slack?

Types of Agents

Slack AI Concierge

Actions
  • Search across workspace.
  • Summaries.
  • Draft replies.
  • Explain content.

Workflow Agents

Actions
  • Trigger workflows.
  • Fill forms.
  • Coordinate multi-step automations.

Enterprise Bots

Actions
  • Pull Salesforce data.
  • Create Jira tickets.
  • Update ServiceNow records.
  • Execute business actions via APIs.

Rules

  • Agents log actions for transparency.
  • Agents cannot access restricted channels.
  • Admins can disable or limit agent capabilities.

🔐 Privacy & Security – AI with a Security Brain

Controls

  • User-level access permissions enforced before AI processing.
  • Encryption (EKM) for enterprise keys.
  • Data never leaves region if residency rules apply.
  • Model choice restricted for sensitive content.

Security Reviews

  • AI models audited.
  • Agent actions reviewed.
  • Workflows approved for automation risk.

🧱 Governance & Guardrails – How You Prevent AI Chaos

Policies

  • AI must be opt-in for enterprise customers.
  • Admins can set visibility rules.
  • No training on customer data unless explicitly allowed.
  • All AI decisions must be reversible.
  • Clear UI labeling for AI-generated content.

Risk Management

  • Hallucination detection.
  • Rate limiting on expensive operations.
  • Fallback to human review for high-impact actions.
  • Sandboxing for third-party AI apps.

📊 AI Performance Metrics – How You Know It’s Working

Reliability

  • Latency for summaries.
  • Search latency impact.
  • Model uptime.

Accuracy

  • Summary correctness.
  • Search ranking quality.
  • Intent classification accuracy.

Adoption

  • Daily summary usage.
  • AI-assisted message drafting.
  • Workflow Builder AI step usage.

Business Outcomes

  • Reduced meeting load.
  • Faster onboarding.
  • Increased cross-team collaboration.

🧑‍💻 Human + AI Collaboration – Division of Labor

Principles
  • AI handles the heavy lifting; humans make final decisions.
  • Users can edit or override any AI output.
  • AI surfaces insights; humans bring judgment.
  • AI explains the “why” behind summaries where possible.
  • Human feedback directly improves future AI performance.

🧙‍♂️ AI Archetype – Who AI “Is” Inside Slack

Guiding question
What is the character of AI inside Slack?
  • Primary archetype: Advisor
  • Secondary archetype: Accelerator
Rationale
Slack AI is a helpful advisor who clarifies information
and an accelerator who executes tasks and automates workflows —
never overshadowing the human, always empowering them.

🧩 How to Use This AI Strand in Practice

  1. Map current AI features
    • List every place AI already appears (search, composer, workflows, bots).
    • Classify them into use cases and surfaces in this strand.
  2. Define your AI architecture + data rules
    • Decide how models are selected,
    • what data they can see,
    • and where guardrails live.
  3. Connect AI to other strands
    • For each Product, UX, Data, Support, and Sales initiative,
    • ask: “What does the AI version of this look like?”
  4. Instrument AI performance
    • Track reliability, accuracy, adoption, and business outcomes.
    • Kill or rework AI that doesn’t move those needles.
  5. Codify human-AI collaboration
    • Make override, feedback, and explainability first-class UX, not afterthoughts.

Screenshotable line:
“Your AI Strand is not about having an AI feature — it’s about turning your entire product into a collaborative intelligence layer.”
{
  "ai_strand": {
    "workshop_meta": {
      "framework_version": "ai-strand-v1.0",
      "source_templates": [
        "AI Purpose",
        "AI Use Cases",
        "AI Surfaces",
        "Model Architecture",
        "Data Inputs",
        "AI Reasoning Abilities",
        "Integration with Product Strands",
        "Automation System",
        "Agents & Bots",
        "Privacy & Security",
        "Governance & Guardrails",
        "AI Performance Metrics",
        "Human + AI Collaboration"
      ],
      "facilitation_notes": [
        "Run with AI/ML, Data, Engineering, Product, Security, Compliance, and Support teams.",
        "Start by reverse-engineering actual AI features in Slack: search, summaries, workflows, automation, agent actions.",
        "This JSON becomes the AI OS — how intelligence is applied everywhere in the company."
      ]
    },

    "purpose_and_role": {
      "question": "Why does AI exist inside Slack?",
      "answer": "AI amplifies human work by summarizing information, accelerating search, automating repetitive tasks, understanding intent, and enabling agents to interact with business systems on behalf of users. AI transforms Slack from a communication tool into a collaborative intelligence layer.",
      "objectives": [
        "Reduce cognitive load in complex, high-velocity communication.",
        "Surface the right information at the right time.",
        "Automate workflows that previously required manual coordination.",
        "Enable teams to work asynchronously without losing context.",
        "Empower users to interact with enterprise systems through natural language."
      ]
    },

    "ai_use_cases": {
      "categories": [
        {
          "name": "Information Understanding",
          "capabilities": [
            "Channel summaries (daily or on-demand)",
            "Thread summaries",
            "Meeting / huddle summaries",
            "Canvas summarization",
            "Long-file summarization"
          ]
        },
        {
          "name": "Search Intelligence",
          "capabilities": [
            "Semantic search powered by embeddings",
            "Search result reranking based on relevance",
            "Federated search across files, messages, and integrated systems",
            "Query rewriting and understanding"
          ]
        },
        {
          "name": "Workflow Intelligence",
          "capabilities": [
            "Suggest workflow automations",
            "Auto-fill workflow steps",
            "Extract structured data from conversations",
            "Trigger actions based on patterns in channels"
          ]
        },
        {
          "name": "Agent Interactions",
          "capabilities": [
            "AI agents that execute tasks (create tickets, update CRM, schedule meetings)",
            "Multi-step reasoning to interact with APIs",
            "Slack-native bot personas and skills",
            "Agents that collaborate with each other in shared channels"
          ]
        },
        {
          "name": "User Assistance",
          "capabilities": [
            "Draft message generation",
            "Rewrite for clarity, tone, conciseness",
            "Explain complex topics",
            "Translate messages",
            "Generate templates (announcements, standups, updates)"
          ]
        },
        {
          "name": "Security & Compliance",
          "capabilities": [
            "Sensitive data detection",
            "Risky user behavior alerts",
            "Automated compliance checks"
          ]
        }
      ]
    },

    "ai_surfaces": {
      "question": "Where does AI appear inside Slack?",
      "surfaces": [
        "Search bar (AI semantic search + summaries)",
        "Message composer (drafts, rewrites, tone changes)",
        "Message actions (summarize thread, summarize channel)",
        "Home tab (AI insights and workflows)",
        "Workflow Builder (AI step suggestions)",
        "Canvas (AI tooling for documentation)",
        "Files (AI summaries of PDFs, docs, images)",
        "AI sidebar (concierge and agent orchestration)"
      ]
    },

    "model_architecture": {
      "question": "How is Slack AI architected?",
      "components": [
        {
          "name": "Slack AI Core",
          "role": "Controls routing, memory, context packaging, and guardrails."
        },
        {
          "name": "LLM Layer",
          "models": [
            "Salesforce Einstein models",
            "Partner LLMs (e.g., OpenAI, Anthropic, Cohere)",
            "Fine-tuned domain-specific models"
          ],
          "notes": "Model selection based on task type, data sensitivity, and cost/latency."
        },
        {
          "name": "Embedding Layer",
          "role": "Generates semantic vectors for search and summarization.",
          "stores": "Vector DB embedded inside Slack search architecture."
        },
        {
          "name": "Context Management",
          "role": "Retrieval, compression, and chunking of messages, threads, and files."
        },
        {
          "name": "AI Execution Engine",
          "role": "Runs agents, multi-step workflows, and action sequences."
        }
      ]
    },

    "data_inputs": {
      "question": "What data does AI use and how is it controlled?",
      "data_sources": [
        "Messages",
        "Threads",
        "Channels",
        "Files",
        "Canvas documents",
        "Lists/projects",
        "Workflow logs",
        "Telemetric signals",
        "User preferences"
      ],
      "privacy_controls": [
        "AI only accesses data users already have permission to view.",
        "Workspace-level admin controls for AI features.",
        "Enterprise key management (EKM) for encryption.",
        "Model and data routing based on compliance region."
      ],
      "excluded_data": [
        "Admin-only messages",
        "Private security logs",
        "Restricted channels unless user is a member"
      ]
    },

    "reasoning_abilities": {
      "question": "What cognitive tasks does Slack AI perform?",
      "abilities": [
        "Summarization",
        "Classification",
        "Semantic retrieval",
        "Intent detection",
        "Context distillation",
        "Step-sequencing for workflows",
        "Natural language to structured logic mapping",
        "Query rewriting"
      ]
    },

    "integration_with_other_strands": {
      "product": [
        "AI features embedded into channels, search, workflows, canvas."
      ],
      "ux": [
        "AI onboarding flows",
        "AI explainability surfaces"
      ],
      "ui": [
        "AI buttons",
        "AI contextual menus",
        "AI sidebar"
      ],
      "data": [
        "Uses semantic vectors stored in search index",
        "Uses message + file corpus"
      ],
      "support": [
        "AI agent for first-tier support",
        "AI suggestion drafts for human agents"
      ],
      "operations": [
        "AI-based incident routing",
        "AI logs for anomaly detection"
      ],
      "sales": [
        "AI generates ROI stories",
        "AI surfaces usage insights for sales reps"
      ]
    },

    "automation_system": {
      "question": "How does AI power automation inside Slack?",
      "architecture": [
        "Workflow Builder + AI (no-code automation creation)",
        "Bots triggered by message patterns",
        "Agent-based automations",
        "Scheduled digests and summaries"
      ],
      "automation_types": [
        "Notifications based on events",
        "Approvals and routing",
        "Incident alerts",
        "Data extraction from conversations",
        "Running scripts or API calls"
      ],
      "governance": [
        "Admin rules for workflows",
        "AI cannot execute destructive actions without confirmation",
        "Rate limiting on automations"
      ]
    },

    "agents_and_bots": {
      "question": "How do AI agents work inside Slack?",
      "types": [
        {
          "name": "Slack AI Concierge",
          "actions": [
            "Search across workspace",
            "Summaries",
            "Draft replies",
            "Explain content"
          ]
        },
        {
          "name": "Workflow Agents",
          "actions": [
            "Trigger workflows",
            "Fill forms",
            "Coordinate multi-step automations"
          ]
        },
        {
          "name": "Enterprise Bots",
          "actions": [
            "Pull Salesforce data",
            "Create Jira tickets",
            "Update ServiceNow records",
            "Execute business actions"
          ]
        }
      ],
      "rules": [
        "Agents log actions for transparency.",
        "Agents cannot access restricted channels.",
        "Admins can disable agent capabilities."
      ]
    },

    "privacy_and_security": {
      "controls": [
        "User-level access permissions enforced pre-AI processing.",
        "Encryption (EKM) for enterprise keys.",
        "Data never leaves region if residency rules apply.",
        "Model choice restricted for sensitive content."
      ],
      "security_reviews": [
        "AI models audited",
        "Agent actions reviewed",
        "Workflows approved for automation risk"
      ]
    },

    "governance_and_guardrails": {
      "policies": [
        "AI must be opt-in for enterprise customers.",
        "Admins can set visibility rules.",
        "No training on customer data unless explicitly allowed.",
        "All AI decisions must be reversible.",
        "Clear UI labeling for AI-generated content."
      ],
      "risk_management": [
        "Hallucination detection",
        "Rate-limiting on expensive operations",
        "Fallback to human review for high-impact actions",
        "Sandboxing for third-party AI apps"
      ]
    },

    "ai_performance_metrics": {
      "reliability": [
        "Latency for summaries",
        "Search latency impact",
        "Model uptime"
      ],
      "accuracy": [
        "Summary correctness",
        "Search ranking quality",
        "Intent classification accuracy"
      ],
      "adoption": [
        "Daily summary usage",
        "AI-assisted message drafting",
        "Workflow Builder AI step usage"
      ],
      "business_outcomes": [
        "Reduced meeting load",
        "Faster onboarding",
        "Increased cross-team collaboration"
      ]
    },

    "human_ai_collaboration": {
      "principles": [
        "AI handles the heavy lifting; humans make final decisions.",
        "Users can edit or override any AI output.",
        "AI surfaces insights; humans bring judgment.",
        "AI explains ‘why’ behind summaries.",
        "Human feedback directly strengthens future AI performance."
      ]
    },

    "ai_archetype": {
      "question": "What is the character of AI inside Slack?",
      "primary_archetype": "Advisor",
      "secondary_archetype": "Accelerator",
      "rationale": "Slack AI is a helpful advisor who clarifies information and an accelerator who executes tasks and automates workflows — never overshadowing the human, always empowering them."
    }
  }
}