Skip to content

AI at Dapper

REFERENCE | DERIVED | Updated 2026-04-08 | Owner: AI Team

How AI works at Dapper Labs — the intelligence architecture, what agents exist, and how to use them.

The Big Picture

Dapper is transitioning from traditional hierarchy to an intelligence-driven organization. The model: every function runs a loop — sense the environment, decide what to do, act, learn from the result. AI agents accelerate these loops from days to hours.

This isn't hypothetical. AI agents currently generate morning briefings, build campaigns, analyze data, and monitor product health. The team wiki you're reading right now is maintained by AI agents.

The Intelligence Architecture

Five layers, each feeding the next:

Layer Function What It Does Example
Sensing Signal ingestion Collects data from Slack, GitHub, Linear, BigQuery, meetings Morning Brief scans all channels
Context Knowledge management Organizes signals into structured knowledge This wiki, KAAOS knowledge base
Evaluation Analysis & scoring Assesses signals against criteria Heimdall data science, health reports
Execution Action generation Produces recommendations or takes autonomous action Campaign builder, reactivation emails
Operations Coordination Orchestrates multiple loops, resolves conflicts KAAOS daemon, loop herding

Autonomy Levels

Not all AI actions are autonomous. Each capability has a defined autonomy level:

Level Description Example
L0 Human does everything, AI assists with information Ad-hoc BigQuery queries
L1 AI drafts, human reviews and approves Morning Brief (AI writes, CEO reviews)
L2 AI acts, human can override Campaign builder (AI generates, PM adjusts)
L3 AI acts autonomously, human monitors Health report posting to Slack
L4 Fully autonomous, human intervention only on exceptions Wiki maintenance, data pipeline monitoring

Most current systems operate at L1-L2. The goal is to push operational tasks toward L3-L4 while keeping strategic decisions at L0-L1.

Current AI Systems

Campaign Builder

What: WYSIWYG editor for creating and managing product campaigns on the Atlas platform. AI-assisted content generation and scheduling.

Who uses it: Product team (Matt, Guy, Sam, Spencer, Jordan)

Status: Demoed to team. Two PRs pending for v1. Moving TST as first production use case.

Heimdall (Data Science Agent)

What: Wraps BigQuery queries to answer data questions in natural language. Can run predefined health scans and generate product metrics.

Who uses it: Data team, product leadership

Key capability: Portfolio health scan across all three products — revenue, segment transitions, pipeline health.

Morning Brief System

What: Replaces traditional standup meetings. AI agent scans Slack, GitHub, Linear, and Calendar overnight, then posts a structured brief in the morning.

Format: Slack thread with progressive disclosure — headline in channel, details in thread replies.

KAAOS Daemon

What: Knowledge management system that runs hourly. Ingests signals from all sources, updates the knowledge base, generates morning dossiers.

Who uses it: CEO and leadership for strategic awareness.

AI Showrunner (In Development)

What: Autonomous experiment loops for Peak Money. Agent designs email copy, sends invites, tracks conversion, designs next experiment based on results.

Who uses it: Peak Money team (Alan Carr leading build)

How to Use Claude Code

Claude Code is the company's primary AI development tool. It's a CLI that connects to Claude (Anthropic's AI) with access to your codebase, tools, and context.

Basics

  • Available on: Terminal (CLI), VS Code, JetBrains, desktop app
  • What it can do: Read/write code, search codebases, run commands, analyze data, generate content
  • Context: It reads your project's CLAUDE.md file for instructions and preferences

For Non-Engineers

You don't need to write code to benefit from AI at Dapper:

  • Campaign content: Ask the campaign builder to generate copy variants
  • Data questions: Ask Heimdall for metrics (when deployed)
  • Meeting prep: The morning brief system surfaces relevant context
  • Documentation: The wiki agents maintain and update product knowledge

For Engineers

  • Claude Code integrates with your development workflow
  • Use /commit for git operations, /review for code review
  • The CLAUDE.md file in each repo defines project-specific instructions
  • AI agents can be dispatched for parallel development tasks

The Morpheus Vision

The long-term goal: Morpheus — an AI coach that every team member can ask questions and get answers grounded in the company's knowledge base.

"How does the Disney submission process work?" → Morpheus reads this wiki and answers.

"What are the top pipeline gaps for NBA?" → Morpheus reads the data insights and answers.

The wiki's completeness determines Morpheus's usefulness. Every gap in this wiki is a question Morpheus can't answer. That's why maintaining and enriching the wiki matters — it's training data for the AI coach.

What This Means for Your Work

  1. Document your domain. If it's in your head but not in the wiki, the AI can't help anyone with it. Write it down.

  2. Use data findings. The 22 validated findings in Data Science Insights should inform every product decision. Reference them by number (F001-F022).

  3. Think in loops. Your work is a loop: sense what's happening, decide what to do, act, learn from the result. AI agents help you run this loop faster. Define what "good" looks like for your domain, and the AI can help you get there.

  4. Flag gaps. If you ask a question and the answer isn't in the wiki or the data, that's a gap worth flagging. The system gets better as we fill gaps.