Case Study • AI Native Engineering • Engineering Operating Model

Agent-Native Development Platform

Enterprise • Private Repository

Executive Summary

Key Metrics

42
Specialized Agents
16
Skill Routers
59
Memory Files
4
Safety Hooks

Orchestration Architecture

A request is routed by a skill to the owning agent, which reads persistent memory and acts on the codebase — all wrapped in a three-layer safety harness.

Sub-Agents Skills File Memory Safety Hooks
Safety Harness — policy · confirmation gate · hard block Request user intent Skill Router 16 skills 42 Sub-Agents 1:1 product surface File Memory 59 files · domain/service Codebase read · edit · verify agent-blueprint meta-scaffold Verify regression

Scope & Ownership

Agents

Skills

Memory

Operating Model & Domain Boundaries

1:1 Surface Mapping

Meta-Scaffolding

Compounding Knowledge

Three-Layer Safety Harness

Layer 1 — Policy

Layer 2 — Confirmation Gate

Layer 3 — Hard Block Hook

Review as Separation of Duties

Organizational Knowledge Layers

Technology Stack

Orchestration
Claude Code Sub-Agents Skills
Knowledge
Markdown Memory Feedback Ledger
Safety
PreToolUse Hooks Permission Gates Time-boxed Access
Quality
Worktree Isolation Parallel Review

Impact

Velocity

Safety

Knowledge

Architecture