AI Engineering Adoption

AI in Production

Before mainstream AI agent frameworks existed,
we were already running multi-agent orchestration in production.
From a LINE bot to Claude Code — a journey from insight to implementation.

2025 — now 20+ Agents 100+ Talks 30+ n8n Flows

Evolution Timeline

2025 / 04

LINE Bot "Little N" Goes Live

"Little N" appeared in the company's LINE group — an AI assistant powered by Flowise + n8n with 20+ specialized agents behind it. From task dispatching and Jenkins deployments to smoke testing and meeting assistance, it covered DevOps, QA, PM, SA, HR, and Finance roles. This was the company's first real AI agent system in production.

2025 / H2

Weekly AI Sharing Sessions

Every Thursday, a dedicated AI knowledge sharing session — from prompt engineering to agent architectures, from local models to cloud services. Over a year, we accumulated 100+ sessions, making it the team's primary channel for learning AI.

2026

The OpenClaw Era

When OpenClaw emerged, we pivoted immediately — because Little N had already proven multi-agent orchestration works. What we upgraded wasn't the idea; it was the infrastructure. Gateway, memory system, local embeddings — all in place.

Now

The Line Between Human and AI is Blurring

Claude Code + Codex CLI + Telegram bridge — dispatching work to a person or to an AI feels increasingly the same. From the LINE bot's semantic routing to Claude Code's autonomous execution, this evolution has been natural.

20+

Specialized Agents

100+

Internal Talks

30+

n8n Automations

6+

Agent Role Categories

Little N's Capability Matrix

Each agent has a clear responsibility, with semantic routing dispatching tasks automatically

Task Dispatcher

Understands user intent semantically and routes to the right specialized agent. The gateway to the entire system.

DevOps / SRE

Jenkins deployment triggers, deployment status queries, server monitoring, UPS power tracking. Deploy with a single message.

QA Testing

Frontend & backend smoke testing, test environment health checks, environment URL lookup.

Project Management

Project progress tracking, requirement visit counts, project database binding and lookup.

SA / Document Search

Template and layout queries per project, vector search across Notion and internal wiki.

Finance / Scheduling / Utilities

Maintenance case status, AR tracking, scheduled reminders, QR code generation, and daily support tasks.

Architecture Evolution

Phase 1 — LINE Bot

Flowise orchestration
n8n automation
Semantic routing

Phase 2 — Sharing

Weekly Thursday sessions
Prompt templates
Cross-team adoption

Phase 3 — OpenClaw

Gateway + memory
Local embeddings
Multi-model switching

Phase 4 — Now

Claude Code + Codex
Telegram bridge
Human-AI boundary blur

The Insight

Looking back, the multi-agent system we built with Flowise + n8n in early 2025
is fundamentally the same thing that LangChain, CrewAI, and OpenClaw later standardized.

The only difference — we started earlier.

"Our insight was precise, our actions were fast, and we had similar concepts before the industry did."
This isn't hindsight. It's the result of continuous experimentation and real-world deployment.